\n\n \n\n \n\n \n \n\n \n\n \n \n\n \n\n \n

\n generated by\n \n $\"bibbase.org\"\n$ \n \n

\n \n\n

\n\n

\n \n Group by\n \n \n
- \n Year\n
- \n Author\n
- \n Type\n
- \n Keyword\n
- \n Downloads\n
\n
\n \n \n \n \n
- \n \n Expand/Collapse All\n \n
- \n \n Download BibTeX\n \n
- \n \n RSS Feed\n \n
\n

\n\n\n

\n\n Excellent! Next you can\n create a new website with this list, or\n embed it in an existing web page by copying & pasting\n any of the following snippets.\n\n

\n JavaScript\n (easiest)\n

\n            <script src=\"https://bibbase.org/show?bib=https%3A%2F%2Fbibbase.org%2Fzotero-group%2Fndrmyrvtl%2F5786256&jsonp=1&jsonp=1\"></script>\n

\n\n PHP\n

\n            <?php\n            $contents = file_get_contents(\"https://bibbase.org/show?bib=https%3A%2F%2Fbibbase.org%2Fzotero-group%2Fndrmyrvtl%2F5786256&jsonp=1\");\n            print_r($contents);\n            ?>\n

\n\n iFrame\n (not recommended)\n

\n            <iframe src=\"https://bibbase.org/show?bib=https%3A%2F%2Fbibbase.org%2Fzotero-group%2Fndrmyrvtl%2F5786256&jsonp=1\"></iframe>\n

\n\n

\n For more details see the documention.\n

\n\n

\n\n This is a preview! To use this list on your own web site\n or create a new web site from it,\n create a free account. The file will be added\n and you will be able to edit it in the File Manager.\n We will show you instructions once you've created your account.\n

\n\n

To the site owner:

\n\n

Action required! Mendeley is changing its\n API. In order to keep using Mendeley with BibBase past April\n 14th, you need to:\n

renew the authorization for BibBase on Mendeley, and
update the BibBase URL\n in your page the same way you did when you initially set up\n this page.\n

\n\n

\n \n \n Fix it now\n

\n\n

\n\n\n

\n \n \n

\n \n 2026\n \n \n (16)\n \n \n

\n \n \n

\n \n\n \n \n \n \n \n \n Multidisciplinary Perspectives on Human‑AI Team Trust.\n \n \n \n \n\n\n \n Tielman, M. L.; Bailey, M.; Frattolillo, F.; Jorge, C. C.; Ulfert, A.; and Meyer-Vitali, A.\n\n\n \n\n\n\n Interaction Studies, 26(2): 164–199. February 2026.\n \n\n\n\n
\n\n\n\n \n \n $\"Multidisciplinary$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{tielmanMultidisciplinaryPerspectivesHumanAI2026,\n\ttitle = {Multidisciplinary {Perspectives} on {Human}‑{AI} {Team} {Trust}},\n\tvolume = {26},\n\tissn = {1572-0373, 1572-0381},\n\turl = {https://www.jbe-platform.com/content/journals/10.1075/is.24048.tie},\n\tdoi = {10.1075/is.24048.tie},\n\tabstract = {Abstract Human-AI teamwork is no longer a topic of the future. Given the importance of trust in human teams, the question arises how trust functions in human-AI teams. Although trust has long been studied from a human-centred perspective (e.g. in psychology and philosophy), a computational perspective and from the perspective of human trust in AI (e.g. in human-computer interaction), the study of trust in human-AI interaction in a team setting is still a novel field. For this reason, the MULTITTRUST (Multidisciplinary perspectives on Human-AI Team Trust) workshop series was founded. In this paper, we present the main outcomes after three editions. Our contributions are: an overview of the shared language of concepts and definitions; an outline of the main open research challenges; and methodological guidelines for further studies in meaningful human-AI team trust. These three contributions form a foundational roadmap towards a better understanding of trust in human-AI team interactions.},\n\tlanguage = {en},\n\tnumber = {2},\n\turldate = {2026-03-02},\n\tjournal = {Interaction Studies},\n\tpublisher = {John Benjamins},\n\tauthor = {Tielman, Myrthe L. and Bailey, Morgan and Frattolillo, Francesco and Jorge, Carolina Centeio and Ulfert, Anna-Sophie and Meyer-Vitali, André},\n\tmonth = feb,\n\tyear = {2026},\n\tpages = {164--199},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n ActivationReasoning: Logical Reasoning in Latent Activation Spaces.\n \n \n \n \n\n\n \n Helff, L.; Härle, R.; Stammer, W.; Friedrich, F.; Brack, M.; Wüst, A.; Shindo, H.; Schramowski, P.; and Kersting, K.\n\n\n \n\n\n\n February 2026.\n arXiv:2510.18184 [cs]\n\n\n\n
\n\n\n\n \n \n $\"ActivationReasoning:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n\n\n\n

@misc{helffActivationReasoningLogicalReasoning2026,\n\ttitle = {{ActivationReasoning}: {Logical} {Reasoning} in {Latent} {Activation} {Spaces}},\n\tshorttitle = {{ActivationReasoning}},\n\turl = {http://arxiv.org/abs/2510.18184},\n\tdoi = {10.48550/arXiv.2510.18184},\n\tabstract = {Large language models (LLMs) excel at generating fluent text, but their internal reasoning remains opaque and difficult to control. Sparse autoencoders (SAEs) make hidden activations more interpretable by exposing latent features that often align with human concepts. Yet, these features are fragile and passive, offering no mechanism for systematic reasoning or model control. To address this, we introduce ActivationReasoning (AR), a framework that embeds explicit logical reasoning into the latent space of LLMs. It proceeds in three stages: (1) Finding latent representations, first latent concept representations are identified (e.g., via SAEs) and organized into a dictionary; (2) Activating propositions, at inference time AR detects activating concepts and maps them to logical propositions; and (3)Logical reasoning, applying logical rules over these propositions to infer higher-order structures, compose new concepts, and steer model behavior. We evaluate AR on multi-hop reasoning (PrOntoQA), abstraction and robustness to indirect concept cues (Rail2Country), reasoning over natural and diverse language (ProverQA), and context-sensitive safety (BeaverTails). Across all tasks, AR scales robustly with reasoning complexity, generalizes to abstract and context-sensitive tasks, and transfers across model backbones. These results demonstrate that grounding logical structure in latent activations not only improves transparency but also enables structured reasoning, reliable control, and alignment with desired behaviors, providing a path toward more reliable and auditable AI.},\n\turldate = {2026-02-26},\n\tpublisher = {arXiv},\n\tauthor = {Helff, Lukas and Härle, Ruben and Stammer, Wolfgang and Friedrich, Felix and Brack, Manuel and Wüst, Antonia and Shindo, Hikaru and Schramowski, Patrick and Kersting, Kristian},\n\tmonth = feb,\n\tyear = {2026},\n\tnote = {arXiv:2510.18184 [cs]},\n\tkeywords = {Computer Science - Artificial Intelligence, Computer Science - Machine Learning},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Let’s Talk AI with Philosophy and Computer Science Expert Kevin Baum.\n \n \n \n \n\n\n \n Baum, K.; and Steffen, B.\n\n\n \n\n\n\n In Steffen, B.; Lee, E. A.; and Steffen, B., editor(s), Let’s Talk AI, volume 15000, pages 104–112. Springer Nature Switzerland, Cham, 2026.\n \n\n\n\n
\n\n\n\n \n \n $\"Let’s$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@incollection{baumLetsTalkAI2026,\n\taddress = {Cham},\n\ttitle = {Let’s {Talk} {AI} with {Philosophy} and {Computer} {Science} {Expert} {Kevin} {Baum}},\n\tvolume = {15000},\n\tisbn = {978-3-032-09007-2 978-3-032-09008-9},\n\turl = {https://link.springer.com/10.1007/978-3-032-09008-9_12},\n\tabstract = {Abstract\n            Building trustworthy AI comes with numerous challenges, ranging from robustness and fairness to explainability for effective human oversight and responsible decision-making. Interdisciplinary collaboration is key for tackling these challenges – fortunately, as the AI community grows, finding shared understanding and common ground between relevant fields becomes easier, because more and more researchers with interdisciplinary backgrounds are entering the field. This paves the way for responsible AI development.\n            My personal AI mission: As a philosopher and computer scientist, I am driven to advance responsible AI development by promoting interdisciplinary dialogue and integrating ethical considerations into core research practices. My mission is to create an environment where appropriate trust assessments in AI becomes the norm, ensuring technology serves humanity in a just and responsible manner.},\n\tlanguage = {en},\n\turldate = {2026-02-26},\n\tbooktitle = {Let’s {Talk} {AI}},\n\tpublisher = {Springer Nature Switzerland},\n\tauthor = {Baum, Kevin and Steffen, Barbara},\n\teditor = {Steffen, Barbara and Lee, Edward A. and Steffen, Bernhard},\n\tyear = {2026},\n\tdoi = {10.1007/978-3-032-09008-9_12},\n\tpages = {104--112},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Justifications for Democratizing AI Alignment and Their Prospects.\n \n \n \n \n\n\n \n Steingrüber, A.; and Baum, K.\n\n\n \n\n\n\n In Steffen, B., editor(s), Bridging the Gap Between AI and Reality, volume 16220, pages 146–159. Springer Nature Switzerland, Cham, 2026.\n \n\n\n\n
\n\n\n\n \n \n $\"Justifications$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@incollection{steingruberJustificationsDemocratizingAI2026,\n\taddress = {Cham},\n\ttitle = {Justifications for {Democratizing} {AI} {Alignment} and {Their} {Prospects}},\n\tvolume = {16220},\n\tisbn = {978-3-032-07131-6 978-3-032-07132-3},\n\turl = {https://link.springer.com/10.1007/978-3-032-07132-3_10},\n\tabstract = {Abstract\n            The AI alignment problem comprises both technical and normative dimensions. While technical solutions focus on implementing normative constraints in AI systems, the normative problem concerns determining what these constraints should be. This paper examines justifications for democratic approaches to the normative problem—where affected stakeholders determine AI alignment—as opposed to epistocratic approaches that defer to normative experts. We analyze both instrumental justifications (democratic approaches produce better outcomes) and non-instrumental justifications (democratic approaches prevent illegitimate authority or coercion). We argue that normative and metanormative uncertainty create a justificatory gap that democratic approaches aim to fill through political rather than theoretical justification. However, we identify significant challenges for democratic approaches, particularly regarding the prevention of illegitimate coercion through AI alignment. Our analysis suggests that neither purely epistocratic nor purely democratic approaches may be sufficient on their own, pointing toward hybrid frameworks that combine expert judgment with participatory input alongside institutional safeguards against AI monopolization.},\n\tlanguage = {en},\n\turldate = {2026-02-26},\n\tbooktitle = {Bridging the {Gap} {Between} {AI} and {Reality}},\n\tpublisher = {Springer Nature Switzerland},\n\tauthor = {Steingrüber, Andre and Baum, Kevin},\n\teditor = {Steffen, Bernhard},\n\tyear = {2026},\n\tdoi = {10.1007/978-3-032-07132-3_10},\n\tpages = {146--159},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Order in the Evaluation Court: A Critical Analysis of NLG Evaluation Trends.\n \n \n \n \n\n\n \n Yang, J.; Feldhus, N.; Mohtaj, S.; Hennig, L.; Wang, Q.; Metheniti, E.; Hakimov, S.; Jakob, C.; Solopova, V.; Rieck, K.; Schlangen, D.; Möller, S.; and Schmitt, V.\n\n\n \n\n\n\n January 2026.\n arXiv:2601.07648 [cs]\n\n\n\n
\n\n\n\n \n \n $\"Order$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@misc{yangOrderEvaluationCourt2026,\n\ttitle = {Order in the {Evaluation} {Court}: {A} {Critical} {Analysis} of {NLG} {Evaluation} {Trends}},\n\tshorttitle = {Order in the {Evaluation} {Court}},\n\turl = {http://arxiv.org/abs/2601.07648},\n\tdoi = {10.48550/arXiv.2601.07648},\n\tabstract = {Despite advances in Natural Language Generation (NLG), evaluation remains challenging. Although various new metrics and LLM-as-a-judge (LaaJ) methods are proposed, human judgment persists as the gold standard. To systematically review how NLG evaluation has evolved, we employ an automatic information extraction scheme to gather key information from NLG papers, focusing on different evaluation methods (metrics, LaaJ and human evaluation). With extracted metadata from 14,171 papers across four major conferences (ACL, EMNLP, NAACL, and INLG) over the past six years, we reveal several critical findings: (1) Task Divergence: While Dialogue Generation demonstrates a rapid shift toward LaaJ ({\\textgreater}40\\% in 2025), Machine Translation remains locked into n-gram metrics, and Question Answering exhibits a substantial decline in the proportion of studies conducting human evaluation. (2) Metric Inertia: Despite the development of semantic metrics, general-purpose metrics (e.g., BLEU, ROUGE) continue to be widely used across tasks without empirical justification, often lacking the discriminative power to distinguish between specific quality criteria. (3) Human-LaaJ Divergence: Our association analysis challenges the assumption that LLMs act as mere proxies for humans; LaaJ and human evaluations prioritize very different signals, and explicit validation is scarce ({\\textless}8\\% of papers comparing the two), with only moderate to low correlation. Based on these observations, we derive practical recommendations to improve the rigor of future NLG evaluation.},\n\turldate = {2026-02-26},\n\tpublisher = {arXiv},\n\tauthor = {Yang, Jing and Feldhus, Nils and Mohtaj, Salar and Hennig, Leonhard and Wang, Qianli and Metheniti, Eleni and Hakimov, Sherzod and Jakob, Charlott and Solopova, Veronika and Rieck, Konrad and Schlangen, David and Möller, Sebastian and Schmitt, Vera},\n\tmonth = jan,\n\tyear = {2026},\n\tnote = {arXiv:2601.07648 [cs]},\n\tkeywords = {Computer Science - Computation and Language},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n SLR: Automated Synthesis for Scalable Logical Reasoning.\n \n \n \n \n\n\n \n Helff, L.; Omar, A.; Friedrich, F.; Wüst, A.; Shindo, H.; Mitchell, R.; Woydt, T.; Schramowski, P.; Stammer, W.; and Kersting, K.\n\n\n \n\n\n\n January 2026.\n arXiv:2506.15787 [cs]\n\n\n\n
\n\n\n\n \n \n $\"SLR:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n\n\n\n

@misc{helffSLRAutomatedSynthesis2026,\n\ttitle = {{SLR}: {Automated} {Synthesis} for {Scalable} {Logical} {Reasoning}},\n\tshorttitle = {{SLR}},\n\turl = {http://arxiv.org/abs/2506.15787},\n\tdoi = {10.48550/arXiv.2506.15787},\n\tabstract = {We introduce SLR, an end-to-end framework for systematic evaluation and training of Large Language Models (LLMs) via Scalable Logical Reasoning. Given a user's task specification, SLR automatically synthesizes (i) an instruction prompt for an inductive reasoning task, (ii) a validation program, executable on model outputs to provide verifiable rewards, and (iii) the latent ground-truth rule. This process is fully automated, scalable, requires no human annotations, and offers precise control over task difficulty. Using SLR, we create SLR-Bench, a benchmark comprising 19k prompts organized into 20 curriculum levels that progressively increase in relational, arithmetic, and recursive complexity. Large-scale evaluation reveals that contemporary LLMs readily produce syntactically valid rules, yet often fail at correct logical inference. Recent reasoning LLMs demonstrate improved performance but incur very high test-time computation, with costs exceeding \\$300 for just 1,000 prompts. Finally, curriculum learning via SLR doubles Llama-3-8B accuracy on SLR-Bench, achieving parity with Gemini-Flash-Thinking at a fraction of computational cost. Moreover, these reasoning capabilities generalize to a wide range of established benchmarks, underscoring the effectiveness of SLR for downstream reasoning.},\n\turldate = {2026-02-26},\n\tpublisher = {arXiv},\n\tauthor = {Helff, Lukas and Omar, Ahmad and Friedrich, Felix and Wüst, Antonia and Shindo, Hikaru and Mitchell, Rupert and Woydt, Tim and Schramowski, Patrick and Stammer, Wolfgang and Kersting, Kristian},\n\tmonth = jan,\n\tyear = {2026},\n\tnote = {arXiv:2506.15787 [cs]},\n\tkeywords = {Computer Science - Artificial Intelligence, Computer Science - Computation and Language, Computer Science - Machine Learning},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Persona Prompting as a Lens on LLM Social Reasoning.\n \n \n \n \n\n\n \n Yang, J.; Hechtbauer, M.; Khalilov, E.; Brinkmann, E. L.; Schmitt, V.; and Feldhus, N.\n\n\n \n\n\n\n January 2026.\n arXiv:2601.20757 [cs]\n\n\n\n
\n\n\n\n \n \n $\"Persona$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@misc{yangPersonaPromptingLens2026,\n\ttitle = {Persona {Prompting} as a {Lens} on {LLM} {Social} {Reasoning}},\n\turl = {http://arxiv.org/abs/2601.20757},\n\tdoi = {10.48550/arXiv.2601.20757},\n\tabstract = {For socially sensitive tasks like hate speech detection, the quality of explanations from Large Language Models (LLMs) is crucial for factors like user trust and model alignment. While Persona prompting (PP) is increasingly used as a way to steer model towards user-specific generation, its effect on model rationales remains underexplored. We investigate how LLM-generated rationales vary when conditioned on different simulated demographic personas. Using datasets annotated with word-level rationales, we measure agreement with human annotations from different demographic groups, and assess the impact of PP on model bias and human alignment. Our evaluation across three LLMs results reveals three key findings: (1) PP improving classification on the most subjective task (hate speech) but degrading rationale quality. (2) Simulated personas fail to align with their real-world demographic counterparts, and high inter-persona agreement shows models are resistant to significant steering. (3) Models exhibit consistent demographic biases and a strong tendency to over-flag content as harmful, regardless of PP. Our findings reveal a critical trade-off: while PP can improve classification in socially-sensitive tasks, it often comes at the cost of rationale quality and fails to mitigate underlying biases, urging caution in its application.},\n\turldate = {2026-02-26},\n\tpublisher = {arXiv},\n\tauthor = {Yang, Jing and Hechtbauer, Moritz and Khalilov, Elisabeth and Brinkmann, Evelyn Luise and Schmitt, Vera and Feldhus, Nils},\n\tmonth = jan,\n\tyear = {2026},\n\tnote = {arXiv:2601.20757 [cs]},\n\tkeywords = {Computer Science - Computation and Language},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Utilitarismus und das Problem kollektiven Handelns.\n \n \n \n \n\n\n \n Baum, K.\n\n\n \n\n\n\n In Handbuch Utilitarismus, pages 163–176. Springer, 2026.\n \n\n\n\n
\n\n\n\n \n \n $\"Utilitarismus$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@incollection{baumUtilitarismusUndProblem2026,\n\ttitle = {Utilitarismus und das {Problem} kollektiven {Handelns}},\n\turl = {https://link.springer.com/chapter/10.1007/978-3-662-71326-6_16},\n\turldate = {2026-02-26},\n\tbooktitle = {Handbuch {Utilitarismus}},\n\tpublisher = {Springer},\n\tauthor = {Baum, Kevin},\n\tyear = {2026},\n\tpages = {163--176},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Can Large Language Models Still Explain Themselves? Investigating the Impact of Quantization on Self-Explanations.\n \n \n \n \n\n\n \n Wang, Q.; Feldhus, N.; Atanasova, P.; Splitt, F.; Ostermann, S.; Möller, S.; and Schmitt, V.\n\n\n \n\n\n\n January 2026.\n arXiv:2601.00282 [cs]\n\n\n\n
\n\n\n\n \n \n $\"Can$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n\n\n\n

@misc{wangCanLargeLanguage2026,\n\ttitle = {Can {Large} {Language} {Models} {Still} {Explain} {Themselves}? {Investigating} the {Impact} of {Quantization} on {Self}-{Explanations}},\n\tshorttitle = {Can {Large} {Language} {Models} {Still} {Explain} {Themselves}?},\n\turl = {http://arxiv.org/abs/2601.00282},\n\tdoi = {10.48550/arXiv.2601.00282},\n\tabstract = {Quantization is widely used to accelerate inference and streamline the deployment of large language models (LLMs), yet its effects on self-explanations (SEs) remain unexplored. SEs, generated by LLMs to justify their own outputs, require reasoning about the model's own decision-making process, a capability that may exhibit particular sensitivity to quantization. As SEs are increasingly relied upon for transparency in high-stakes applications, understanding whether and to what extent quantization degrades SE quality and faithfulness is critical. To address this gap, we examine two types of SEs: natural language explanations (NLEs) and counterfactual examples, generated by LLMs quantized using three common techniques at distinct bit widths. Our findings indicate that quantization typically leads to moderate declines in both SE quality (up to 4.4{\\textbackslash}\\%) and faithfulness (up to 2.38{\\textbackslash}\\%). The user study further demonstrates that quantization diminishes both the coherence and trustworthiness of SEs (up to 8.5{\\textbackslash}\\%). Compared to smaller models, larger models show limited resilience to quantization in terms of SE quality but better maintain faithfulness. Moreover, no quantization technique consistently excels across task accuracy, SE quality, and faithfulness. Given that quantization's impact varies by context, we recommend validating SE quality for specific use cases, especially for NLEs, which show greater sensitivity. Nonetheless, the relatively minor deterioration in SE quality and faithfulness does not undermine quantization's effectiveness as a model compression technique.},\n\turldate = {2026-01-23},\n\tpublisher = {arXiv},\n\tauthor = {Wang, Qianli and Feldhus, Nils and Atanasova, Pepa and Splitt, Fedor and Ostermann, Simon and Möller, Sebastian and Schmitt, Vera},\n\tmonth = jan,\n\tyear = {2026},\n\tnote = {arXiv:2601.00282 [cs]},\n\tkeywords = {Computer Science - Artificial Intelligence, Computer Science - Computation and Language, Computer Science - Machine Learning},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Breaking Up with Normatively Monolithic Agency with GRACE: A Reason-Based Neuro-Symbolic Architecture for Safe and Ethical AI Alignment.\n \n \n \n \n\n\n \n Jahn, F.; Muskalla, Y.; Dargasz, L.; Schramowski, P.; and Baum, K.\n\n\n \n\n\n\n January 2026.\n arXiv:2601.10520 [cs]\n\n\n\n
\n\n\n\n \n \n $\"Breaking$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n\n\n\n

@misc{jahnBreakingNormativelyMonolithic2026,\n\ttitle = {Breaking {Up} with {Normatively} {Monolithic} {Agency} with {GRACE}: {A} {Reason}-{Based} {Neuro}-{Symbolic} {Architecture} for {Safe} and {Ethical} {AI} {Alignment}},\n\tshorttitle = {Breaking {Up} with {Normatively} {Monolithic} {Agency} with {GRACE}},\n\turl = {http://arxiv.org/abs/2601.10520},\n\tdoi = {10.48550/arXiv.2601.10520},\n\tabstract = {As AI agents become increasingly autonomous, widely deployed in consequential contexts, and efficacious in bringing about real-world impacts, ensuring that their decisions are not only instrumentally effective but also normatively aligned has become critical. We introduce a neuro-symbolic reason-based containment architecture, Governor for Reason-Aligned ContainmEnt (GRACE), that decouples normative reasoning from instrumental decision-making and can contain AI agents of virtually any design. GRACE restructures decision-making into three modules: a Moral Module (MM) that determines permissible macro actions via deontic logic-based reasoning; a Decision-Making Module (DMM) that encapsulates the target agent while selecting instrumentally optimal primitive actions in accordance with derived macro actions; and a Guard that monitors and enforces moral compliance. The MM uses a reason-based formalism providing a semantic foundation for deontic logic, enabling interpretability, contestability, and justifiability. Its symbolic representation enriches the DMM's informational context and supports formal verification and statistical guarantees of alignment enforced by the Guard. We demonstrate GRACE on an example of a LLM therapy assistant, showing how it enables stakeholders to understand, contest, and refine agent behavior.},\n\turldate = {2026-01-23},\n\tpublisher = {arXiv},\n\tauthor = {Jahn, Felix and Muskalla, Yannic and Dargasz, Lisa and Schramowski, Patrick and Baum, Kevin},\n\tmonth = jan,\n\tyear = {2026},\n\tnote = {arXiv:2601.10520 [cs]},\n\tkeywords = {Computer Science - Artificial Intelligence, Computer Science - Computers and Society},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n CLaS-Bench: A Cross-Lingual Alignment and Steering Benchmark.\n \n \n \n \n\n\n \n Gurgurov, D.; Ghussin, Y. A.; Baeumel, T.; Chou, C.; Schramowski, P.; Mosbach, M.; Genabith, J. v.; and Ostermann, S.\n\n\n \n\n\n\n January 2026.\n arXiv:2601.08331 [cs]\n\n\n\n
\n\n\n\n \n \n $\"CLaS-Bench:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@misc{gurgurovCLaSBenchCrossLingualAlignment2026,\n\ttitle = {{CLaS}-{Bench}: {A} {Cross}-{Lingual} {Alignment} and {Steering} {Benchmark}},\n\tshorttitle = {{CLaS}-{Bench}},\n\turl = {http://arxiv.org/abs/2601.08331},\n\tdoi = {10.48550/arXiv.2601.08331},\n\tabstract = {Understanding and controlling the behavior of large language models (LLMs) is an increasingly important topic in multilingual NLP. Beyond prompting or fine-tuning, , i.e.,{\\textasciitilde}manipulating internal representations during inference, has emerged as a more efficient and interpretable technique for adapting models to a target language. Yet, no dedicated benchmarks or evaluation protocols exist to quantify the effectiveness of steering techniques. We introduce CLaS-Bench, a lightweight parallel-question benchmark for evaluating language-forcing behavior in LLMs across 32 languages, enabling systematic evaluation of multilingual steering methods. We evaluate a broad array of steering techniques, including residual-stream DiffMean interventions, probe-derived directions, language-specific neurons, PCA/LDA vectors, Sparse Autoencoders, and prompting baselines. Steering performance is measured along two axes: language control and semantic relevance, combined into a single harmonic-mean steering score. We find that across languages simple residual-based DiffMean method consistently outperforms all other methods. Moreover, a layer-wise analysis reveals that language-specific structure emerges predominantly in later layers and steering directions cluster based on language family. CLaS-Bench is the first standardized benchmark for multilingual steering, enabling both rigorous scientific analysis of language representations and practical evaluation of steering as a low-cost adaptation alternative.},\n\turldate = {2026-01-23},\n\tpublisher = {arXiv},\n\tauthor = {Gurgurov, Daniil and Ghussin, Yusser Al and Baeumel, Tanja and Chou, Cheng-Ting and Schramowski, Patrick and Mosbach, Marius and Genabith, Josef van and Ostermann, Simon},\n\tmonth = jan,\n\tyear = {2026},\n\tnote = {arXiv:2601.08331 [cs]},\n\tkeywords = {Computer Science - Computation and Language},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n On the Complexities of Testing for Compliance with Human Oversight Requirements in AI Regulation.\n \n \n \n \n\n\n \n Langer, M.; Lazar, V.; and Baum, K.\n\n\n \n\n\n\n In Steffen, B., editor(s), Bridging the Gap Between AI and Reality, volume 16220, pages 160–169. Springer Nature Switzerland, Cham, 2026.\n \n\n\n\n
\n\n\n\n \n \n $\"On$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@incollection{langerComplexitiesTestingCompliance2026,\n\taddress = {Cham},\n\ttitle = {On the {Complexities} of {Testing} for {Compliance} with {Human} {Oversight} {Requirements} in {AI} {Regulation}},\n\tvolume = {16220},\n\tisbn = {978-3-032-07131-6 978-3-032-07132-3},\n\turl = {https://link.springer.com/10.1007/978-3-032-07132-3_11},\n\tdoi = {10.1007/978-3-032-07132-3_11},\n\tabstract = {Abstract\n            Human oversight requirements are a core component of the European AI Act and in AI governance. In this paper, we highlight key challenges in testing for compliance with these requirements. A central difficulty lies in balancing simple, but potentially ineffective checklist-based approaches with resource-intensive and context-sensitive empirical testing of the effectiveness of human oversight of AI. Questions regarding when to update compliance testing, the context-dependent nature of human oversight requirements, and difficult-to-operationalize standards further complicate compliance testing. We argue that these challenges illustrate broader challenges in the future of sociotechnical AI governance, i.e. a future that shifts from ensuring “good” technological products to “good” sociotechnical systems.},\n\tlanguage = {en},\n\turldate = {2026-01-23},\n\tbooktitle = {Bridging the {Gap} {Between} {AI} and {Reality}},\n\tpublisher = {Springer Nature Switzerland},\n\tauthor = {Langer, Markus and Lazar, Veronika and Baum, Kevin},\n\teditor = {Steffen, Bernhard},\n\tyear = {2026},\n\tpages = {160--169},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n From Weights to Activations: Is Steering the Next Frontier of Adaptation?.\n \n \n \n \n\n\n \n Ostermann, S.; Gurgurov, D.; Baeumel, T.; Hedderich, M. A.; Lapuschkin, S.; Samek, W.; and Schmitt, V.\n\n\n \n\n\n\n . January 2026.\n \n\n\n\n
\n\n\n\n \n \n $\"From$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@article{ostermannWeightsActivationsSteering2026,\n\ttitle = {From {Weights} to {Activations}: {Is} {Steering} the {Next} {Frontier} of {Adaptation}?},\n\tshorttitle = {From {Weights} to {Activations}},\n\turl = {https://www.researchgate.net/profile/Simon-Ostermann-6/publication/399528610_From_Weights_to_Activations_Is_Steering_the_Next_Frontier_of_Adaptation/links/695e221b7e61d05b5317e780/From-Weights-to-Activations-Is-Steering-the-Next-Frontier-of-Adaptation.pdf},\n\turldate = {2026-01-23},\n\tauthor = {Ostermann, Simon and Gurgurov, Daniil and Baeumel, Tanja and Hedderich, Michael A. and Lapuschkin, Sebastian and Samek, Wojciech and Schmitt, Vera},\n\tmonth = jan,\n\tyear = {2026},\n\tkeywords = {⛔ No DOI found},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Through a Compressed Lens: Investigating The Impact of Quantization on Factual Knowledge Recall.\n \n \n \n \n\n\n \n Wang, Q.; Wang, M.; Feldhus, N.; Ostermann, S.; Cao, Y.; Schütze, H.; Möller, S.; and Schmitt, V.\n\n\n \n\n\n\n January 2026.\n arXiv:2505.13963 [cs]\n\n\n\n
\n\n\n\n \n \n $\"Through$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n\n\n\n

@misc{wangCompressedLensInvestigating2026,\n\ttitle = {Through a {Compressed} {Lens}: {Investigating} {The} {Impact} of {Quantization} on {Factual} {Knowledge} {Recall}},\n\tshorttitle = {Through a {Compressed} {Lens}},\n\turl = {http://arxiv.org/abs/2505.13963},\n\tdoi = {10.48550/arXiv.2505.13963},\n\tabstract = {Quantization methods are widely used to accelerate inference and streamline the deployment of large language models (LLMs). Although quantization's effects on various LLM capabilities have been extensively studied, one critical area remains underexplored: factual knowledge recall (FKR), the process by which LLMs access stored knowledge. To this end, we conduct comprehensive experiments using three common quantization techniques at distinct bit widths, in conjunction with interpretability-driven analyses on two tasks, knowledge memorization and latent multi-hop reasoning. We show that quantization typically results in information loss within LLMs, consequently diminishing their capacity for FKR. This effect is particularly amplified in smaller models within the same architectural families. However, models quantized at reduced bit precision do not consistently exhibit inferior performance and occasionally quantization may even enhance model FKR. We find that BitSandBytes demonstrates highest preservation of the original full-precision model's FKR. Despite variability across models and methods, quantization causes modest performance degradation and remains an effective compression strategy.},\n\turldate = {2026-01-23},\n\tpublisher = {arXiv},\n\tauthor = {Wang, Qianli and Wang, Mingyang and Feldhus, Nils and Ostermann, Simon and Cao, Yuan and Schütze, Hinrich and Möller, Sebastian and Schmitt, Vera},\n\tmonth = jan,\n\tyear = {2026},\n\tnote = {arXiv:2505.13963 [cs]},\n\tkeywords = {Computer Science - Computation and Language, Computer Science - Machine Learning},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Task Prompt Vectors: Effective Initialization Through Multi-task Soft Prompt Transfer.\n \n \n \n \n\n\n \n Belanec, R.; Ostermann, S.; Srba, I.; and Bielikova, M.\n\n\n \n\n\n\n In Pfahringer, B.; Japkowicz, N.; Larrañaga, P.; Ribeiro, R. P.; Dutra, I.; Pechenizkiy, M.; Cortez, P.; Pashami, S.; Jorge, A. M.; Soares, C.; Abreu, P. H.; and Gama, J., editor(s), Machine Learning and Knowledge Discovery in Databases. Research Track and Applied Data Science Track, volume 16020, pages 77–94. Springer Berlin Heidelberg, Berlin, Heidelberg, 2026.\n \n\n\n\n
\n\n\n\n \n \n $\"Task$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@incollection{belanecTaskPromptVectors2026,\n\taddress = {Berlin, Heidelberg},\n\ttitle = {Task {Prompt} {Vectors}: {Effective} {Initialization} {Through} {Multi}-task {Soft} {Prompt} {Transfer}},\n\tvolume = {16020},\n\tisbn = {978-3-662-72242-8 978-3-662-72243-5},\n\tshorttitle = {Task {Prompt} {Vectors}},\n\turl = {https://link.springer.com/10.1007/978-3-662-72243-5_5},\n\tdoi = {10.1007/978-3-662-72243-5_5},\n\tlanguage = {en},\n\turldate = {2026-01-23},\n\tbooktitle = {Machine {Learning} and {Knowledge} {Discovery} in {Databases}. {Research} {Track} and {Applied} {Data} {Science} {Track}},\n\tpublisher = {Springer Berlin Heidelberg},\n\tauthor = {Belanec, Robert and Ostermann, Simon and Srba, Ivan and Bielikova, Maria},\n\teditor = {Pfahringer, Bernhard and Japkowicz, Nathalie and Larrañaga, Pedro and Ribeiro, Rita P. and Dutra, Inês and Pechenizkiy, Mykola and Cortez, Paulo and Pashami, Sepideh and Jorge, Alípio M. and Soares, Carlos and Abreu, Pedro H. and Gama, João},\n\tyear = {2026},\n\tpages = {77--94},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Responsible and Trusted AI: An Interdisciplinary Perspective (2025).\n \n \n \n \n\n\n \n Kerstan, S.; Baum, K.; Helfer, T.; Langer, M.; Schmidt, E.; Sesing-Wagenpfeil, A.; and Speith, T.\n\n\n \n\n\n\n In Steffen, B., editor(s), Bridging the Gap Between AI and Reality, volume 16220, pages 141–145. Springer Nature Switzerland, Cham, 2026.\n \n\n\n\n
\n\n\n\n \n \n $\"Responsible$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@incollection{kerstanResponsibleTrustedAI2026,\n\taddress = {Cham},\n\ttitle = {Responsible and {Trusted} {AI}: {An} {Interdisciplinary} {Perspective} (2025)},\n\tvolume = {16220},\n\tisbn = {978-3-032-07131-6 978-3-032-07132-3},\n\tshorttitle = {Responsible and {Trusted} {AI}},\n\turl = {https://link.springer.com/10.1007/978-3-032-07132-3_9},\n\tdoi = {10.1007/978-3-032-07132-3_9},\n\tabstract = {Abstract\n            As Artificial Intelligence (AI) continues to shape individual lives, institutional processes, and societal structures, ensuring its responsible and trusted development has become a critical imperative. However, meeting this imperative is far from straightforward. AI systems frequently lack transparency and are embedded in environments where the distribution of responsibility and accountability is unclear, normative standards are disputed, and system behavior is unpredictable. The Responsible and Trusted AI track at AISoLA 2025 addresses these and similar challenges by fostering interdisciplinary collaboration across philosophy, law, psychology, economics, sociology, political science, and informatics. This introduction outlines the motivation for the track, emphasizing the sociotechnical embeddedness of AI and the need for approaches that go beyond technical performance to consider questions related to trust and responsibility. It highlights three core themes explored in this year’s contributions: democratic legitimation and normative alignment, legal compliance and human oversight, and runtime safety in high-risk contexts. Together, these contributions underscore the importance of interdisciplinary discussions to navigate normative ambiguity, regulatory uncertainty, and behavioral unpredictability in AI systems. The track aims to advance dialogue and collaboration that support the development and deployment of AI systems that are not only effective but are also designed and implemented responsibly and can be trusted.},\n\tlanguage = {en},\n\turldate = {2026-01-23},\n\tbooktitle = {Bridging the {Gap} {Between} {AI} and {Reality}},\n\tpublisher = {Springer Nature Switzerland},\n\tauthor = {Kerstan, Sophie and Baum, Kevin and Helfer, Thorsten and Langer, Markus and Schmidt, Eva and Sesing-Wagenpfeil, Andreas and Speith, Timo},\n\teditor = {Steffen, Bernhard},\n\tyear = {2026},\n\tpages = {141--145},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n

\n \n 2025\n \n \n (78)\n \n \n

\n \n \n

\n \n\n \n \n \n \n \n \n Sortability of Time Series Data.\n \n \n \n \n\n\n \n Lohse, C.; and Wahl, J.\n\n\n \n\n\n\n August 2025.\n arXiv:2407.13313 [cs]\n\n\n\n
\n\n\n\n \n \n $\"Sortability$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@misc{lohseSortabilityTimeSeries2025,\n\ttitle = {Sortability of {Time} {Series} {Data}},\n\turl = {http://arxiv.org/abs/2407.13313},\n\tdoi = {10.48550/arXiv.2407.13313},\n\tabstract = {Evaluating the performance of causal discovery algorithms that aim to find causal relationships between time-dependent processes remains a challenging topic. In this paper, we show that certain characteristics of datasets, such as varsortability (Reisach et al. 2021) and \\$R{\\textasciicircum}2\\$-sortability (Reisach et al. 2023), also occur in datasets for autocorrelated stationary time series. We illustrate this empirically using four types of data: simulated data based on SVAR models and Erdős-Rényi graphs, the data used in the 2019 causality-for-climate challenge (Runge et al. 2019), real-world river stream datasets, and real-world data generated by the Causal Chamber of (Gamella et al. 2024). To do this, we adapt var- and \\$R{\\textasciicircum}2\\$-sortability to time series data. We also investigate the extent to which the performance of score-based causal discovery methods goes hand in hand with high sortability. Arguably, our most surprising finding is that the investigated real-world datasets exhibit high varsortability and low \\$R{\\textasciicircum}2\\$-sortability indicating that scales may carry a significant amount of causal information.},\n\turldate = {2026-02-26},\n\tpublisher = {arXiv},\n\tauthor = {Lohse, Christopher and Wahl, Jonas},\n\tmonth = aug,\n\tyear = {2025},\n\tnote = {arXiv:2407.13313 [cs]},\n\tkeywords = {Computer Science - Artificial Intelligence},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n When Counterfactual Reasoning Fails: Chaos and Real-World Complexity.\n \n \n \n \n\n\n \n Aalaila, Y.; Großmann, G.; Mukherjee, S.; Wahl, J.; and Vollmer, S.\n\n\n \n\n\n\n April 2025.\n arXiv:2503.23820 [cs]\n\n\n\n
\n\n\n\n \n \n $\"When$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n\n\n\n

@misc{aalailaWhenCounterfactualReasoning2025,\n\ttitle = {When {Counterfactual} {Reasoning} {Fails}: {Chaos} and {Real}-{World} {Complexity}},\n\tshorttitle = {When {Counterfactual} {Reasoning} {Fails}},\n\turl = {http://arxiv.org/abs/2503.23820},\n\tdoi = {10.48550/arXiv.2503.23820},\n\tabstract = {Counterfactual reasoning, a cornerstone of human cognition and decision-making, is often seen as the 'holy grail' of causal learning, with applications ranging from interpreting machine learning models to promoting algorithmic fairness. While counterfactual reasoning has been extensively studied in contexts where the underlying causal model is well-defined, real-world causal modeling is often hindered by model and parameter uncertainty, observational noise, and chaotic behavior. The reliability of counterfactual analysis in such settings remains largely unexplored. In this work, we investigate the limitations of counterfactual reasoning within the framework of Structural Causal Models. Specifically, we empirically investigate {\\textbackslash}emph\\{counterfactual sequence estimation\\} and highlight cases where it becomes increasingly unreliable. We find that realistic assumptions, such as low degrees of model uncertainty or chaotic dynamics, can result in counterintuitive outcomes, including dramatic deviations between predicted and true counterfactual trajectories. This work urges caution when applying counterfactual reasoning in settings characterized by chaos and uncertainty. Furthermore, it raises the question of whether certain systems may pose fundamental limitations on the ability to answer counterfactual questions about their behavior.},\n\turldate = {2026-02-26},\n\tpublisher = {arXiv},\n\tauthor = {Aalaila, Yahya and Großmann, Gerrit and Mukherjee, Sumantrak and Wahl, Jonas and Vollmer, Sebastian},\n\tmonth = apr,\n\tyear = {2025},\n\tnote = {arXiv:2503.23820 [cs]},\n\tkeywords = {Computer Science - Artificial Intelligence, Computer Science - Machine Learning},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Aggregation Problems in Machine Ethics and AI Alignment.\n \n \n \n \n\n\n \n Baum, K.; and Slavkovik, M.\n\n\n \n\n\n\n In Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, volume 8, pages 355–366, 2025. \n \n\n\n\n
\n\n\n\n \n \n $\"Aggregation$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{baumAggregationProblemsMachine2025,\n\ttitle = {Aggregation {Problems} in {Machine} {Ethics} and {AI} {Alignment}},\n\tvolume = {8},\n\turl = {https://ojs.aaai.org/index.php/AIES/article/view/36554},\n\turldate = {2026-02-26},\n\tbooktitle = {Proceedings of the {AAAI}/{ACM} {Conference} on {AI}, {Ethics}, and {Society}},\n\tauthor = {Baum, Kevin and Slavkovik, Marija},\n\tyear = {2025},\n\tpages = {355--366},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n A Road Map for Responsible Robotics: Promoting Human Agency and Collaborative Efforts.\n \n \n \n \n\n\n \n Araiza-Illan, D.; Baum, K.; Beebee, H.; Chatila, R.; Christensen, S. M.; Coghlan, S.; Collins, E.; Conroy, S.; Cunha, A.; Dobrosovestnova, A.; Duijf, H.; Evers, V.; Fisher, M.; Hochgeschwender, N.; Kökciyan, N.; Lemaignan, S.; Rodriguez-Lera, F.; Ljungblad, S.; Magnusson, M.; Mansouri, M.; Milford, M.; Moon, A.; Powers, T. M.; Salvini, P.; Scantamburlo, T.; Schuster, N.; Slavkovik, M.; Topcu, U.; Vanegas, D.; Wasowski, A.; and Yang, Y.\n\n\n \n\n\n\n IEEE Robotics & Automation Magazine, 32(4): 12–24. December 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"A$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n\n\n\n

@article{araiza-illanRoadMapResponsible2025,\n\ttitle = {A {Road} {Map} for {Responsible} {Robotics}: {Promoting} {Human} {Agency} and {Collaborative} {Efforts}},\n\tvolume = {32},\n\tissn = {1558-223X},\n\tshorttitle = {A {Road} {Map} for {Responsible} {Robotics}},\n\turl = {https://ieeexplore.ieee.org/document/11236462/},\n\tdoi = {10.1109/MRA.2025.3620148},\n\tabstract = {This document presents the outcomes of the Dagstuhl Seminar “Roadmap for Responsible Robotics,” held in September 2023 at the Leibniz Center for Informatics, Schloss Dagstuhl, Germany. The seminar brought together researchers from the fields of robotics, computer science, social and cognitive sciences, and philosophy with the aim of charting a path toward improving responsibility in robotic systems. Through intensive interdisciplinary discussions centered on the various values at stake as robotics increasingly integrates into human life, the participants identified key priorities to guide future research and regulatory efforts. The resulting roadmap outlines actionable steps to ensure that robotic systems coevolve with human societies, promoting human agency and humane values rather than undermining them. Designed for diverse stakeholders—researchers, policy makers, industry leaders, practitioners, nongovernmental organizations (NGOs), and civil society groups—this roadmap provides a foundation for collaborative efforts toward responsible robotics.},\n\tnumber = {4},\n\turldate = {2026-02-26},\n\tjournal = {IEEE Robotics \\& Automation Magazine},\n\tauthor = {Araiza-Illan, Dejanira and Baum, Kevin and Beebee, Helen and Chatila, Raja and Christensen, Sarah Moth-Lund and Coghlan, Simon and Collins, Emily and Conroy, S.Kate and Cunha, Alcino and Dobrosovestnova, Anna and Duijf, Hein and Evers, Vanessa and Fisher, Michael and Hochgeschwender, Nico and Kökciyan, Nadin and Lemaignan, Séverin and Rodriguez-Lera, Francisco and Ljungblad, Sara and Magnusson, Martin and Mansouri, Masoumeh and Milford, Michael and Moon, AJung and Powers, Thomas M. and Salvini, Pericle and Scantamburlo, Teresa and Schuster, Nick and Slavkovik, Marija and Topcu, Ufuk and Vanegas, Daniel and Wasowski, Andrzej and Yang, Yi},\n\tmonth = dec,\n\tyear = {2025},\n\tkeywords = {Automation, Collaboration, Computer science, Industries, Informatics, Non-governmental organizations, Philosophical considerations, Roads, Seminars, Service robots},\n\tpages = {12--24},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n The Landscape of Causal Discovery Data: Grounding Causal Discovery in Real-World Applications.\n \n \n \n \n\n\n \n Brouillard, P.; Squires, C.; Wahl, J.; Kording, K. P.; Sachs, K.; Drouin, A.; and Sridhar, D.\n\n\n \n\n\n\n June 2025.\n arXiv:2412.01953 [cs]\n\n\n\n
\n\n\n\n \n \n $\"The$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n\n\n\n

@misc{brouillardLandscapeCausalDiscovery2025,\n\ttitle = {The {Landscape} of {Causal} {Discovery} {Data}: {Grounding} {Causal} {Discovery} in {Real}-{World} {Applications}},\n\tshorttitle = {The {Landscape} of {Causal} {Discovery} {Data}},\n\turl = {http://arxiv.org/abs/2412.01953},\n\tdoi = {10.48550/arXiv.2412.01953},\n\tabstract = {Causal discovery aims to automatically uncover causal relationships from data, a capability with significant potential across many scientific disciplines. However, its real-world applications remain limited. Current methods often rely on unrealistic assumptions and are evaluated only on simple synthetic toy datasets, often with inadequate evaluation metrics. In this paper, we substantiate these claims by performing a systematic review of the recent causal discovery literature. We present applications in biology, neuroscience, and Earth sciences - fields where causal discovery holds promise for addressing key challenges. We highlight available simulated and real-world datasets from these domains and discuss common assumption violations that have spurred the development of new methods. Our goal is to encourage the community to adopt better evaluation practices by utilizing realistic datasets and more adequate metrics.},\n\turldate = {2026-02-26},\n\tpublisher = {arXiv},\n\tauthor = {Brouillard, Philippe and Squires, Chandler and Wahl, Jonas and Kording, Konrad P. and Sachs, Karen and Drouin, Alexandre and Sridhar, Dhanya},\n\tmonth = jun,\n\tyear = {2025},\n\tnote = {arXiv:2412.01953 [cs]},\n\tkeywords = {Computer Science - Machine Learning, Statistics - Methodology},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Beyond Overcorrection: Evaluating Diversity in T2I Models with DivBench.\n \n \n \n \n\n\n \n Friedrich, F.; Welsch, T. G.; Brack, M.; Schramowski, P.; and Kersting, K.\n\n\n \n\n\n\n July 2025.\n arXiv:2507.03015 [cs]\n\n\n\n
\n\n\n\n \n \n $\"Beyond$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n\n\n\n

@misc{friedrichOvercorrectionEvaluatingDiversity2025,\n\ttitle = {Beyond {Overcorrection}: {Evaluating} {Diversity} in {T2I} {Models} with {DivBench}},\n\tshorttitle = {Beyond {Overcorrection}},\n\turl = {http://arxiv.org/abs/2507.03015},\n\tdoi = {10.48550/arXiv.2507.03015},\n\tabstract = {Current diversification strategies for text-to-image (T2I) models often ignore contextual appropriateness, leading to over-diversification where demographic attributes are modified even when explicitly specified in prompts. This paper introduces DIVBENCH, a benchmark and evaluation framework for measuring both under- and over-diversification in T2I generation. Through systematic evaluation of state-of-the-art T2I models, we find that while most models exhibit limited diversity, many diversification approaches overcorrect by inappropriately altering contextually-specified attributes. We demonstrate that context-aware methods, particularly LLM-guided FairDiffusion and prompt rewriting, can already effectively address under-diversity while avoiding over-diversification, achieving a better balance between representation and semantic fidelity.},\n\turldate = {2026-02-26},\n\tpublisher = {arXiv},\n\tauthor = {Friedrich, Felix and Welsch, Thiemo Ganesha and Brack, Manuel and Schramowski, Patrick and Kersting, Kristian},\n\tmonth = jul,\n\tyear = {2025},\n\tnote = {arXiv:2507.03015 [cs]},\n\tkeywords = {Computer Science - Computation and Language, Computer Science - Computers and Society, Computer Science - Machine Learning},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n ART: Adaptive Relation Tuning for Generalized Relation Prediction.\n \n \n \n \n\n\n \n Sudhakaran, G.; Shindo, H.; Schramowski, P.; Schaub-Meyer, S.; Kersting, K.; and Roth, S.\n\n\n \n\n\n\n In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 16323–16332, 2025. \n \n\n\n\n
\n\n\n\n \n \n $\"ART:$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{sudhakaranARTAdaptiveRelation2025,\n\ttitle = {{ART}: {Adaptive} {Relation} {Tuning} for {Generalized} {Relation} {Prediction}},\n\tshorttitle = {{ART}},\n\turl = {https://openaccess.thecvf.com/content/ICCV2025/html/Sudhakaran_ART_Adaptive_Relation_Tuning_for_Generalized_Relation_Prediction_ICCV_2025_paper.html},\n\turldate = {2026-02-26},\n\tbooktitle = {Proceedings of the {IEEE}/{CVF} {International} {Conference} on {Computer} {Vision}},\n\tauthor = {Sudhakaran, Gopika and Shindo, Hikaru and Schramowski, Patrick and Schaub-Meyer, Simone and Kersting, Kristian and Roth, Stefan},\n\tyear = {2025},\n\tpages = {16323--16332},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n AILuminate: Introducing v1.0 of the AI Risk and Reliability Benchmark from MLCommons.\n \n \n \n \n\n\n \n Ghosh, S.; Frase, H.; Williams, A.; Luger, S.; Röttger, P.; Barez, F.; McGregor, S.; Fricklas, K.; Kumar, M.; Feuillade–Montixi, Q.; Bollacker, K.; Friedrich, F.; Tsang, R.; Vidgen, B.; Parrish, A.; Knotz, C.; Presani, E.; Bennion, J.; Boston, M. F.; Kuniavsky, M.; Hutiri, W.; Ezick, J.; Salem, M. B.; Sahay, R.; Goswami, S.; Gohar, U.; Huang, B.; Sarin, S.; Alhajjar, E.; Chen, C.; Eng, R.; Manjusha, K. R.; Mehta, V.; Long, E.; Emani, M.; Vidra, N.; Rukundo, B.; Shahbazi, A.; Chen, K.; Ghosh, R.; Thangarasa, V.; Peigné, P.; Singh, A.; Bartolo, M.; Krishna, S.; Akhtar, M.; Gold, R.; Coleman, C.; Oala, L.; Tashev, V.; Imperial, J. M.; Russ, A.; Kunapuli, S.; Miailhe, N.; Delaunay, J.; Radharapu, B.; Shinde, R.; Tuesday; Dutta, D.; Grabb, D.; Gangavarapu, A.; Sahay, S.; Gangavarapu, A.; Schramowski, P.; Singam, S.; David, T.; Han, X.; Mammen, P. M.; Prabhakar, T.; Kovatchev, V.; Weiss, R.; Ahmed, A.; Manyeki, K. N.; Madireddy, S.; Khomh, F.; Zhdanov, F.; Baumann, J.; Vasan, N.; Yang, X.; Mougn, C.; Varghese, J. R.; Chinoy, H.; Jitendar, S.; Maskey, M.; Hardgrove, C. V.; Li, T.; Gupta, A.; Joswin, E.; Mai, Y.; Kumar, S. H.; Patlak, C.; Lu, K.; Alessi, V.; Balija, S. B.; Gu, C.; Sullivan, R.; Gealy, J.; Lavrisa, M.; Goel, J.; Mattson, P.; Liang, P.; and Vanschoren, J.\n\n\n \n\n\n\n April 2025.\n arXiv:2503.05731 [cs]\n\n\n\n
\n\n\n\n \n \n $\"AILuminate:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n\n\n\n

@misc{ghoshAILuminateIntroducingV102025,\n\ttitle = {{AILuminate}: {Introducing} v1.0 of the {AI} {Risk} and {Reliability} {Benchmark} from {MLCommons}},\n\tshorttitle = {{AILuminate}},\n\turl = {http://arxiv.org/abs/2503.05731},\n\tdoi = {10.48550/arXiv.2503.05731},\n\tabstract = {The rapid advancement and deployment of AI systems have created an urgent need for standard safety-evaluation frameworks. This paper introduces AILuminate v1.0, the first comprehensive industry-standard benchmark for assessing AI-product risk and reliability. Its development employed an open process that included participants from multiple fields. The benchmark evaluates an AI system's resistance to prompts designed to elicit dangerous, illegal, or undesirable behavior in 12 hazard categories, including violent crimes, nonviolent crimes, sex-related crimes, child sexual exploitation, indiscriminate weapons, suicide and self-harm, intellectual property, privacy, defamation, hate, sexual content, and specialized advice (election, financial, health, legal). Our method incorporates a complete assessment standard, extensive prompt datasets, a novel evaluation framework, a grading and reporting system, and the technical as well as organizational infrastructure for long-term support and evolution. In particular, the benchmark employs an understandable five-tier grading scale (Poor to Excellent) and incorporates an innovative entropy-based system-response evaluation. In addition to unveiling the benchmark, this report also identifies limitations of our method and of building safety benchmarks generally, including evaluator uncertainty and the constraints of single-turn interactions. This work represents a crucial step toward establishing global standards for AI risk and reliability evaluation while acknowledging the need for continued development in areas such as multiturn interactions, multimodal understanding, coverage of additional languages, and emerging hazard categories. Our findings provide valuable insights for model developers, system integrators, and policymakers working to promote safer AI deployment.},\n\turldate = {2026-02-26},\n\tpublisher = {arXiv},\n\tauthor = {Ghosh, Shaona and Frase, Heather and Williams, Adina and Luger, Sarah and Röttger, Paul and Barez, Fazl and McGregor, Sean and Fricklas, Kenneth and Kumar, Mala and Feuillade--Montixi, Quentin and Bollacker, Kurt and Friedrich, Felix and Tsang, Ryan and Vidgen, Bertie and Parrish, Alicia and Knotz, Chris and Presani, Eleonora and Bennion, Jonathan and Boston, Marisa Ferrara and Kuniavsky, Mike and Hutiri, Wiebke and Ezick, James and Salem, Malek Ben and Sahay, Rajat and Goswami, Sujata and Gohar, Usman and Huang, Ben and Sarin, Supheakmungkol and Alhajjar, Elie and Chen, Canyu and Eng, Roman and Manjusha, Kashyap Ramanandula and Mehta, Virendra and Long, Eileen and Emani, Murali and Vidra, Natan and Rukundo, Benjamin and Shahbazi, Abolfazl and Chen, Kongtao and Ghosh, Rajat and Thangarasa, Vithursan and Peigné, Pierre and Singh, Abhinav and Bartolo, Max and Krishna, Satyapriya and Akhtar, Mubashara and Gold, Rafael and Coleman, Cody and Oala, Luis and Tashev, Vassil and Imperial, Joseph Marvin and Russ, Amy and Kunapuli, Sasidhar and Miailhe, Nicolas and Delaunay, Julien and Radharapu, Bhaktipriya and Shinde, Rajat and {Tuesday} and Dutta, Debojyoti and Grabb, Declan and Gangavarapu, Ananya and Sahay, Saurav and Gangavarapu, Agasthya and Schramowski, Patrick and Singam, Stephen and David, Tom and Han, Xudong and Mammen, Priyanka Mary and Prabhakar, Tarunima and Kovatchev, Venelin and Weiss, Rebecca and Ahmed, Ahmed and Manyeki, Kelvin N. and Madireddy, Sandeep and Khomh, Foutse and Zhdanov, Fedor and Baumann, Joachim and Vasan, Nina and Yang, Xianjun and Mougn, Carlos and Varghese, Jibin Rajan and Chinoy, Hussain and Jitendar, Seshakrishna and Maskey, Manil and Hardgrove, Claire V. and Li, Tianhao and Gupta, Aakash and Joswin, Emil and Mai, Yifan and Kumar, Shachi H. and Patlak, Cigdem and Lu, Kevin and Alessi, Vincent and Balija, Sree Bhargavi and Gu, Chenhe and Sullivan, Robert and Gealy, James and Lavrisa, Matt and Goel, James and Mattson, Peter and Liang, Percy and Vanschoren, Joaquin},\n\tmonth = apr,\n\tyear = {2025},\n\tnote = {arXiv:2503.05731 [cs]},\n\tkeywords = {Computer Science - Artificial Intelligence, Computer Science - Computers and Society},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n CHRONOBERG: Capturing Language Evolution and Temporal Awareness in Foundation Models.\n \n \n \n \n\n\n \n Hegde, N.; Paul, S.; Joel-Frey, L.; Brack, M.; Kersting, K.; Mundt, M.; and Schramowski, P.\n\n\n \n\n\n\n September 2025.\n arXiv:2509.22360 [cs]\n\n\n\n
\n\n\n\n \n \n $\"CHRONOBERG:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n\n\n\n

@misc{hegdeCHRONOBERGCapturingLanguage2025,\n\ttitle = {{CHRONOBERG}: {Capturing} {Language} {Evolution} and {Temporal} {Awareness} in {Foundation} {Models}},\n\tshorttitle = {{CHRONOBERG}},\n\turl = {http://arxiv.org/abs/2509.22360},\n\tdoi = {10.48550/arXiv.2509.22360},\n\tabstract = {Large language models (LLMs) excel at operating at scale by leveraging social media and various data crawled from the web. Whereas existing corpora are diverse, their frequent lack of long-term temporal structure may however limit an LLM's ability to contextualize semantic and normative evolution of language and to capture diachronic variation. To support analysis and training for the latter, we introduce CHRONOBERG, a temporally structured corpus of English book texts spanning 250 years, curated from Project Gutenberg and enriched with a variety of temporal annotations. First, the edited nature of books enables us to quantify lexical semantic change through time-sensitive Valence-Arousal-Dominance (VAD) analysis and to construct historically calibrated affective lexicons to support temporally grounded interpretation. With the lexicons at hand, we demonstrate a need for modern LLM-based tools to better situate their detection of discriminatory language and contextualization of sentiment across various time-periods. In fact, we show how language models trained sequentially on CHRONOBERG struggle to encode diachronic shifts in meaning, emphasizing the need for temporally aware training and evaluation pipelines, and positioning CHRONOBERG as a scalable resource for the study of linguistic change and temporal generalization. Disclaimer: This paper includes language and display of samples that could be offensive to readers. Open Access: Chronoberg is available publicly on HuggingFace at ( https://huggingface.co/datasets/spaul25/Chronoberg). Code is available at (https://github.com/paulsubarna/Chronoberg).},\n\turldate = {2026-02-26},\n\tpublisher = {arXiv},\n\tauthor = {Hegde, Niharika and Paul, Subarnaduti and Joel-Frey, Lars and Brack, Manuel and Kersting, Kristian and Mundt, Martin and Schramowski, Patrick},\n\tmonth = sep,\n\tyear = {2025},\n\tnote = {arXiv:2509.22360 [cs]},\n\tkeywords = {Computer Science - Artificial Intelligence, Computer Science - Computation and Language},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Comparing LLMs and BERT-based Classifiers for Resource-Sensitive Claim Verification in Social Media.\n \n \n \n \n\n\n \n Upravitelev, M.; Duran-Silva, N.; Woerle, C.; Guarino, G.; Mohtaj, S.; Yang, J.; Solopova, V.; and Schmitt, V.\n\n\n \n\n\n\n In Ghosal, T.; Mayr, P.; Singh, A.; Naik, A.; Rehm, G.; Freitag, D.; Li, D.; Schimmler, S.; and De Waard, A., editor(s), Proceedings of the Fifth Workshop on Scholarly Document Processing (SDP 2025), pages 281–287, Vienna, Austria, July 2025. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"Comparing$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{upravitelevComparingLLMsBERTbased2025,\n\taddress = {Vienna, Austria},\n\ttitle = {Comparing {LLMs} and {BERT}-based {Classifiers} for {Resource}-{Sensitive} {Claim} {Verification} in {Social} {Media}},\n\tisbn = {979-8-89176-265-7},\n\turl = {https://aclanthology.org/2025.sdp-1.26/},\n\tdoi = {10.18653/v1/2025.sdp-1.26},\n\tabstract = {The overwhelming volume of content being published at any given moment poses a significant challenge for the design of automated fact-checking (AFC) systems on social media, requiring an emphasized consideration of efficiency aspects.As in other fields, systems built upon LLMs have achieved good results on different AFC benchmarks. However, the application of LLMs is accompanied by high resource requirements. The energy consumption of LLMs poses a significant challenge from an ecological perspective, while remaining a bottleneck in latency-sensitive scenarios like AFC within social media. Therefore, we propose a system built upon fine-tuned smaller BERT-based models. When evaluated on the ClimateCheck dataset against decoder-only LLMs, our best fine-tuned model outperforms Phi 4 and approaches Qwen3 14B in reasoning mode — while significantly reducing runtime per claim. Our findings demonstrate that small encoder-only models fine-tuned for specific tasks can still provide a substantive alternative to large decoder-only LLMs, especially in efficiency-concerned settings.},\n\turldate = {2026-02-26},\n\tbooktitle = {Proceedings of the {Fifth} {Workshop} on {Scholarly} {Document} {Processing} ({SDP} 2025)},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Upravitelev, Max and Duran-Silva, Nicolau and Woerle, Christian and Guarino, Giuseppe and Mohtaj, Salar and Yang, Jing and Solopova, Veronika and Schmitt, Vera},\n\teditor = {Ghosal, Tirthankar and Mayr, Philipp and Singh, Amanpreet and Naik, Aakanksha and Rehm, Georg and Freitag, Dayne and Li, Dan and Schimmler, Sonja and De Waard, Anita},\n\tmonth = jul,\n\tyear = {2025},\n\tpages = {281--287},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Building Common Ground in Dialogue: A Survey.\n \n \n \n \n\n\n \n Anikina, T.; Leippert, A.; and Ostermann, S.\n\n\n \n\n\n\n In Proceedings of the 2nd LUHME Workshop, pages 3–28, 2025. \n \n\n\n\n
\n\n\n\n \n \n $\"Building$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{anikinaBuildingCommonGround2025a,\n\ttitle = {Building {Common} {Ground} in {Dialogue}: {A} {Survey}},\n\tshorttitle = {Building {Common} {Ground} in {Dialogue}},\n\turl = {https://aclanthology.org/2025.luhme-1.2/},\n\turldate = {2026-02-26},\n\tbooktitle = {Proceedings of the 2nd {LUHME} {Workshop}},\n\tauthor = {Anikina, Tatiana and Leippert, Alina and Ostermann, Simon},\n\tyear = {2025},\n\tpages = {3--28},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Multilingual Political Views of Large Language Models: Identification and Steering.\n \n \n \n \n\n\n \n Gurgurov, D.; Trinley, K.; Vykopal, I.; Genabith, J. V.; Ostermann, S.; and Zamparelli, R.\n\n\n \n\n\n\n In Inui, K.; Sakti, S.; Wang, H.; Wong, D. F.; Bhattacharyya, P.; Banerjee, B.; Ekbal, A.; Chakraborty, T.; and Singh, D. P., editor(s), Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics, pages 279–298, Mumbai, India, December 2025. The Asian Federation of Natural Language Processing and The Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"Multilingual$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{gurgurovMultilingualPoliticalViews2025,\n\taddress = {Mumbai, India},\n\ttitle = {Multilingual {Political} {Views} of {Large} {Language} {Models}: {Identification} and {Steering}},\n\tisbn = {979-8-89176-303-6},\n\tshorttitle = {Multilingual {Political} {Views} of {Large} {Language} {Models}},\n\turl = {https://aclanthology.org/2025.findings-ijcnlp.17/},\n\tabstract = {Large language models (LLMs) are increasingly used in everyday tools and applications, raising concerns about their potential influence on political views. While prior research has shown that LLMs often exhibit measurable political biases-frequently skewing toward liberal or progressive positions-key gaps remain. Most existing studies evaluate only a narrow set of models and languages, leaving open questions about the generalizability of political biases across architectures, scales, and multilingual settings. Moreover, few works examine whether these biases can be actively controlled.In this work, we address these gaps through a large-scale study of political orientation in modern open-source instruction-tuned LLMs. We evaluate seven models, including {\\textbackslash}textttLLaMA-3.1, {\\textbackslash}textttQwen-3, and {\\textbackslash}textttAya-Expanse, across 14 languages using the Political Compass Test with 11 semantically equivalent paraphrases per statement to ensure robust measurement. Our results reveal that larger models consistently shift toward libertarian-left positions, with significant variations across languages and model families. To test the manipulability of political stances, we utilize a simple center-of-mass activation intervention technique and show that it reliably steers model responses toward alternative ideological positions across multiple languages. Our code is publicly available at https://github.com/d-gurgurov/Political-Ideologies-LLMs.},\n\turldate = {2026-02-26},\n\tbooktitle = {Proceedings of the 14th {International} {Joint} {Conference} on {Natural} {Language} {Processing} and the 4th {Conference} of the {Asia}-{Pacific} {Chapter} of the {Association} for {Computational} {Linguistics}},\n\tpublisher = {The Asian Federation of Natural Language Processing and The Association for Computational Linguistics},\n\tauthor = {Gurgurov, Daniil and Trinley, Katharina and Vykopal, Ivan and Genabith, Josef Van and Ostermann, Simon and Zamparelli, Roberto},\n\teditor = {Inui, Kentaro and Sakti, Sakriani and Wang, Haofen and Wong, Derek F. and Bhattacharyya, Pushpak and Banerjee, Biplab and Ekbal, Asif and Chakraborty, Tanmoy and Singh, Dhirendra Pratap},\n\tmonth = dec,\n\tyear = {2025},\n\tpages = {279--298},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n LlavaGuard: An Open VLM-based Framework for Safeguarding Vision Datasets and Models.\n \n \n \n \n\n\n \n Helff, L.; Friedrich, F.; Brack, M.; Kersting, K.; and Schramowski, P.\n\n\n \n\n\n\n June 2025.\n arXiv:2406.05113 [cs]\n\n\n\n
\n\n\n\n \n \n $\"LlavaGuard:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n\n\n\n

@misc{helffLlavaGuardOpenVLMbased2025,\n\ttitle = {{LlavaGuard}: {An} {Open} {VLM}-based {Framework} for {Safeguarding} {Vision} {Datasets} and {Models}},\n\tshorttitle = {{LlavaGuard}},\n\turl = {http://arxiv.org/abs/2406.05113},\n\tdoi = {10.48550/arXiv.2406.05113},\n\tabstract = {This paper introduces LlavaGuard, a suite of VLM-based vision safeguards that address the critical need for reliable guardrails in the era of large-scale data and models. To this end, we establish a novel open framework, describing a customizable safety taxonomy, data preprocessing, augmentation, and training setup. For teaching a VLM safeguard on safety, we further create a multimodal safety dataset with high-quality human expert annotations, where each image is labeled with a safety rating, category, and rationale. We also employ advanced augmentations to support context-specific assessments. The resulting LlavaGuard models, ranging from 0.5B to 7B, serve as a versatile tool for evaluating the safety compliance of visual content against flexible policies. In comprehensive experiments, LlavaGuard outperforms both state-of-the-art safeguards and VLMs in accuracy and in flexibly handling different policies. Additionally, we demonstrate LlavaGuard's performance in two real-world applications: large-scale dataset annotation and moderation of text-to-image models. We make our entire framework, including the dataset, model weights, and training code.},\n\turldate = {2026-02-26},\n\tpublisher = {arXiv},\n\tauthor = {Helff, Lukas and Friedrich, Felix and Brack, Manuel and Kersting, Kristian and Schramowski, Patrick},\n\tmonth = jun,\n\tyear = {2025},\n\tnote = {arXiv:2406.05113 [cs]},\n\tkeywords = {Computer Science - Artificial Intelligence, Computer Science - Computer Vision and Pattern Recognition, Computer Science - Machine Learning},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n LIME: Making LLM Data More Efficient with Linguistic Metadata Embeddings.\n \n \n \n \n\n\n \n Sztwiertnia, S.; Friedrich, F.; Kersting, K.; Schramowski, P.; and Deiseroth, B.\n\n\n \n\n\n\n December 2025.\n arXiv:2512.07522 [cs]\n\n\n\n
\n\n\n\n \n \n $\"LIME:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n\n\n\n

@misc{sztwiertniaLIMEMakingLLM2025,\n\ttitle = {{LIME}: {Making} {LLM} {Data} {More} {Efficient} with {Linguistic} {Metadata} {Embeddings}},\n\tshorttitle = {{LIME}},\n\turl = {http://arxiv.org/abs/2512.07522},\n\tdoi = {10.48550/arXiv.2512.07522},\n\tabstract = {Pre-training decoder-only language models relies on vast amounts of high-quality data, yet the availability of such data is increasingly reaching its limits. While metadata is commonly used to create and curate these datasets, its potential as a direct training signal remains under-explored. We challenge this status quo and propose LIME (Linguistic Metadata Embeddings), a method that enriches token embeddings with metadata capturing syntax, semantics, and contextual properties. LIME substantially improves pre-training efficiency. Specifically, it adapts up to 56\\% faster to the training data distribution, while introducing only 0.01\\% additional parameters at negligible compute overhead. Beyond efficiency, LIME improves tokenization, leading to remarkably stronger language modeling capabilities and generative task performance. These benefits persist across model scales (500M to 2B). In addition, we develop a variant with shifted metadata, LIME+1, that can guide token generation. Given prior metadata for the next token, LIME+1 improves reasoning performance by up to 38\\% and arithmetic accuracy by up to 35\\%.},\n\turldate = {2026-02-26},\n\tpublisher = {arXiv},\n\tauthor = {Sztwiertnia, Sebastian and Friedrich, Felix and Kersting, Kristian and Schramowski, Patrick and Deiseroth, Björn},\n\tmonth = dec,\n\tyear = {2025},\n\tnote = {arXiv:2512.07522 [cs]},\n\tkeywords = {Computer Science - Artificial Intelligence, Computer Science - Computation and Language},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Learning Per-Domain Generalizing Policies Using Offline Reinforcement Learning.\n \n \n \n \n\n\n \n Müller, N. J.; Oster, M.; and Gros, T. P.\n\n\n \n\n\n\n . 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"Learning$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{mullerLearningPerDomainGeneralizing2025,\n\ttitle = {Learning {Per}-{Domain} {Generalizing} {Policies} {Using} {Offline} {Reinforcement} {Learning}},\n\turl = {https://prl-theworkshop.github.io/prl2025-icaps/papers/11.pdf},\n\turldate = {2026-02-26},\n\tauthor = {Müller, Nicola J. and Oster, Moritz and Gros, Timo P.},\n\tyear = {2025},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Bridging the Gap Between AI Planning and Reinforcement Learning.\n \n \n \n \n\n\n \n Ajanovi, Z.; Gros, T.; Den Hengst, F.; Holler, D.; Kokel, H.; and Taitler, A.\n\n\n \n\n\n\n In AAAI Conference on Artificial Intelligence, 2025. \n \n\n\n\n
\n\n\n\n \n \n $\"Bridging$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{ajanoviBridgingGapAI2025,\n\ttitle = {Bridging the {Gap} {Between} {AI} {Planning} and {Reinforcement} {Learning}},\n\turl = {https://research.ibm.com/publications/bridging-the-gap-between-ai-planning-and-reinforcement-learning},\n\turldate = {2026-02-26},\n\tbooktitle = {{AAAI} {Conference} on {Artificial} {Intelligence}},\n\tauthor = {Ajanovi, Zlatan and Gros, Timo and Den Hengst, Floris and Holler, Daniel and Kokel, Harsha and Taitler, Ayal},\n\tyear = {2025},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Measuring and Guiding Monosemanticity.\n \n \n \n \n\n\n \n Härle, R.; Friedrich, F.; Brack, M.; Wäldchen, S.; Deiseroth, B.; Schramowski, P.; and Kersting, K.\n\n\n \n\n\n\n December 2025.\n arXiv:2506.19382 [cs]\n\n\n\n
\n\n\n\n \n \n $\"Measuring$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@misc{harleMeasuringGuidingMonosemanticity2025,\n\ttitle = {Measuring and {Guiding} {Monosemanticity}},\n\turl = {http://arxiv.org/abs/2506.19382},\n\tdoi = {10.48550/arXiv.2506.19382},\n\tabstract = {There is growing interest in leveraging mechanistic interpretability and controllability to better understand and influence the internal dynamics of large language models (LLMs). However, current methods face fundamental challenges in reliably localizing and manipulating feature representations. Sparse Autoencoders (SAEs) have recently emerged as a promising direction for feature extraction at scale, yet they, too, are limited by incomplete feature isolation and unreliable monosemanticity. To systematically quantify these limitations, we introduce Feature Monosemanticity Score (FMS), a novel metric to quantify feature monosemanticity in latent representation. Building on these insights, we propose Guided Sparse Autoencoders (G-SAE), a method that conditions latent representations on labeled concepts during training. We demonstrate that reliable localization and disentanglement of target concepts within the latent space improve interpretability, detection of behavior, and control. Specifically, our evaluations on toxicity detection, writing style identification, and privacy attribute recognition show that G-SAE not only enhances monosemanticity but also enables more effective and fine-grained steering with less quality degradation. Our findings provide actionable guidelines for measuring and advancing mechanistic interpretability and control of LLMs.},\n\turldate = {2026-02-26},\n\tpublisher = {arXiv},\n\tauthor = {Härle, Ruben and Friedrich, Felix and Brack, Manuel and Wäldchen, Stephan and Deiseroth, Björn and Schramowski, Patrick and Kersting, Kristian},\n\tmonth = dec,\n\tyear = {2025},\n\tnote = {arXiv:2506.19382 [cs]},\n\tkeywords = {Computer Science - Computation and Language},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Judging Quality Across Languages: A Multilingual Approach to Pretraining Data Filtering with Language Models.\n \n \n \n \n\n\n \n Ali, M.; Brack, M.; Lübbering, M.; Wendt, E.; Khan, A. G.; Rutmann, R.; Jude, A.; Kraus, M.; Weber, A. A.; and Stollenwerk, F.\n\n\n \n\n\n\n In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 8870–8909, 2025. \n \n\n\n\n
\n\n\n\n \n \n $\"Judging$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{aliJudgingQualityLanguages2025,\n\ttitle = {Judging {Quality} {Across} {Languages}: {A} {Multilingual} {Approach} to {Pretraining} {Data} {Filtering} with {Language} {Models}},\n\tshorttitle = {Judging {Quality} {Across} {Languages}},\n\turl = {https://aclanthology.org/2025.emnlp-main.449/},\n\turldate = {2026-02-26},\n\tbooktitle = {Proceedings of the 2025 {Conference} on {Empirical} {Methods} in {Natural} {Language} {Processing}},\n\tauthor = {Ali, Mehdi and Brack, Manuel and Lübbering, Max and Wendt, Elias and Khan, Abbas Goher and Rutmann, Richard and Jude, Alex and Kraus, Maurice and Weber, Alexander Arno and Stollenwerk, Felix},\n\tyear = {2025},\n\tpages = {8870--8909},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n How to Train your Text-to-Image Model: Evaluating Design Choices for Synthetic Training Captions.\n \n \n \n \n\n\n \n Brack, M.; Katakol, S.; Friedrich, F.; Schramowski, P.; Ravi, H.; Kersting, K.; and Kale, A.\n\n\n \n\n\n\n In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 6823–6832, 2025. \n \n\n\n\n
\n\n\n\n \n \n $\"How$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{brackHowTrainYour2025,\n\ttitle = {How to {Train} your {Text}-to-{Image} {Model}: {Evaluating} {Design} {Choices} for {Synthetic} {Training} {Captions}},\n\tshorttitle = {How to {Train} your {Text}-to-{Image} {Model}},\n\turl = {https://openaccess.thecvf.com/content/ICCV2025W/CDEL/html/Brack_How_to_Train_your_Text-to-Image_Model_Evaluating_Design_Choices_for_ICCVW_2025_paper.html},\n\turldate = {2026-02-26},\n\tbooktitle = {Proceedings of the {IEEE}/{CVF} {International} {Conference} on {Computer} {Vision}},\n\tauthor = {Brack, Manuel and Katakol, Sudeep and Friedrich, Felix and Schramowski, Patrick and Ravi, Hareesh and Kersting, Kristian and Kale, Ajinkya},\n\tyear = {2025},\n\tpages = {6823--6832},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Doing Wrong with Others: Multi-Agent Consequentialism as a Solution for the Collective Action Problem.\n \n \n \n \n\n\n \n Baum, K.\n\n\n \n\n\n\n . 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"Doing$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{baumDoingWrongOthers2025,\n\ttitle = {Doing {Wrong} with {Others}: {Multi}-{Agent} {Consequentialism} as a {Solution} for the {Collective} {Action} {Problem}},\n\tshorttitle = {Doing {Wrong} with {Others}},\n\turl = {https://philarchive.org/rec/BAUDWW},\n\turldate = {2026-02-26},\n\tauthor = {Baum, Kevin},\n\tyear = {2025},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Cross-Prompt Encoder for Low-Performing Languages.\n \n \n \n \n\n\n \n Mikaberidze, B.; Saghinadze, T.; Ostermann, S.; and Müller, P.\n\n\n \n\n\n\n In Inui, K.; Sakti, S.; Wang, H.; Wong, D. F.; Bhattacharyya, P.; Banerjee, B.; Ekbal, A.; Chakraborty, T.; and Singh, D. P., editor(s), Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics, pages 2380–2393, Mumbai, India, December 2025. The Asian Federation of Natural Language Processing and The Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"Cross-Prompt$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{mikaberidzeCrossPromptEncoderLowPerforming2025,\n\taddress = {Mumbai, India},\n\ttitle = {Cross-{Prompt} {Encoder} for {Low}-{Performing} {Languages}},\n\tisbn = {979-8-89176-303-6},\n\turl = {https://aclanthology.org/2025.findings-ijcnlp.144/},\n\tabstract = {Soft prompts have emerged as a powerful alternative to adapters in parameter-efficient fine-tuning (PEFT), enabling large language models (LLMs) to adapt to downstream tasks without architectural changes or parameter updates. While prior work has focused on stabilizing training via parameter interaction in small neural prompt encoders, their broader potential for transfer across languages remains unexplored. In this paper, we demonstrate that a prompt encoder can play a central role in improving performance on low-performing languages—those that achieve poor accuracy even under full-model fine-tuning. We investigate a lightweight encoder paired with multi-source training on typologically diverse languages. We call this architecture-training combination the Cross-Prompt Encoder (XPE), and show that it advances the capture of abstract, transferable patterns across languages. To complement XPE, we propose a Dual Soft Prompt mechanism that combines an encoder-based prompt with a directly trained standard soft prompt. This hybrid design proves especially effective for target languages that benefit from both broadly shared structure and language-specific alignment. Text classification experiments with a transformer encoder (XLM-R) on the SIB-200 benchmark reveal a consistent trade-off: XPE is most effective for low-performing languages, while hybrid variants offer broader adaptability across multilingual settings.},\n\turldate = {2026-02-26},\n\tbooktitle = {Proceedings of the 14th {International} {Joint} {Conference} on {Natural} {Language} {Processing} and the 4th {Conference} of the {Asia}-{Pacific} {Chapter} of the {Association} for {Computational} {Linguistics}},\n\tpublisher = {The Asian Federation of Natural Language Processing and The Association for Computational Linguistics},\n\tauthor = {Mikaberidze, Beso and Saghinadze, Temo and Ostermann, Simon and Müller, Philipp},\n\teditor = {Inui, Kentaro and Sakti, Sakriani and Wang, Haofen and Wong, Derek F. and Bhattacharyya, Pushpak and Banerjee, Biplab and Ekbal, Asif and Chakraborty, Tanmoy and Singh, Dhirendra Pratap},\n\tmonth = dec,\n\tyear = {2025},\n\tpages = {2380--2393},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Multilingual Datasets for Custom Input Extraction and Explanation Requests Parsing in Conversational XAI Systems.\n \n \n \n \n\n\n \n Wang, Q.; Anikina, T.; Feldhus, N.; Ostermann, S.; Splitt, F.; Li, J.; Tsoneva, Y.; Möller, S.; and Schmitt, V.\n\n\n \n\n\n\n arXiv preprint arXiv:2508.14982. 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"Multilingual$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{wangMultilingualDatasetsCustom2025a,\n\ttitle = {Multilingual {Datasets} for {Custom} {Input} {Extraction} and {Explanation} {Requests} {Parsing} in {Conversational} {XAI} {Systems}},\n\turl = {https://aclanthology.org/anthology-files/anthology-files/pdf/findings/2025.findings-emnlp.29.pdf},\n\turldate = {2026-02-26},\n\tjournal = {arXiv preprint arXiv:2508.14982},\n\tauthor = {Wang, Qianli and Anikina, Tatiana and Feldhus, Nils and Ostermann, Simon and Splitt, Fedor and Li, Jiaao and Tsoneva, Yoana and Möller, Sebastian and Schmitt, Vera},\n\tyear = {2025},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Looking for Something Causal? Die versteckten Beziehungen der Daten.\n \n \n \n \n\n\n \n Baum, K.\n\n\n \n\n\n\n IM+ io, (3). 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"Looking$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{baumLookingSomethingCausal2025,\n\ttitle = {Looking for {Something} {Causal}? {Die} versteckten {Beziehungen} der {Daten}.},\n\tshorttitle = {Looking for {Something} {Causal}?},\n\turl = {https://search.ebscohost.com/login.aspx?direct=true&profile=ehost&scope=site&authtype=crawler&jrnl=21989990&asa=N&AN=188665449&h=d6cxwd2ZBlowJXnoUfUpGfuVS0SdotS%2BmLzH5qksZ7y52Bas3HKkrlevJ0VoiCWV%2F28iTdikHOQXia4loAxUTQ%3D%3D&crl=c},\n\tnumber = {3},\n\turldate = {2026-02-26},\n\tjournal = {IM+ io},\n\tauthor = {Baum, Kevin},\n\tyear = {2025},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Per-Domain Generalizing Policies for Classical Planning: On Scaling Behavior and Validation Instances.\n \n \n \n \n\n\n \n Gros, T. P.; Müller, N. J.; Fišer, D.; Valera, I.; Wolf, V.; and Hoffmann, J.\n\n\n \n\n\n\n . 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"Per-Domain$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{grosPerDomainGeneralizingPolicies2025a,\n\ttitle = {Per-{Domain} {Generalizing} {Policies} for {Classical} {Planning}: {On} {Scaling} {Behavior} and {Validation} {Instances}},\n\tshorttitle = {Per-{Domain} {Generalizing} {Policies} for {Classical} {Planning}},\n\turl = {https://mosi.uni-saarland.de/assets/papers/genplan-AAAI25.pdf},\n\turldate = {2026-02-26},\n\tauthor = {Gros, Timo P. and Müller, Nicola J. and Fišer, Daniel and Valera, Isabel and Wolf, Verena and Hoffmann, Jörg},\n\tyear = {2025},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Per-domain generalizing policies: On validation instances and scaling behavior.\n \n \n \n \n\n\n \n Gros, T. P.; Müller, N. J.; Fišer, D.; Valera, I.; Wolf, V.; and Hoffmann, J.\n\n\n \n\n\n\n In Proceedings of the International Conference on Automated Planning and Scheduling, volume 35, pages 198–203, 2025. \n \n\n\n\n
\n\n\n\n \n \n $\"Per-domain$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{grosPerdomainGeneralizingPolicies2025,\n\ttitle = {Per-domain generalizing policies: {On} validation instances and scaling behavior},\n\tvolume = {35},\n\tshorttitle = {Per-domain generalizing policies},\n\turl = {https://ojs.aaai.org/index.php/ICAPS/article/view/36118},\n\turldate = {2026-02-26},\n\tbooktitle = {Proceedings of the {International} {Conference} on {Automated} {Planning} and {Scheduling}},\n\tauthor = {Gros, Timo P. and Müller, Nicola J. and Fišer, Daniel and Valera, Isabel and Wolf, Verena and Hoffmann, Jörg},\n\tyear = {2025},\n\tpages = {198--203},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n SIG.\n \n \n \n \n\n\n \n Estevanell-Valladares, E. L.; Picazo-Izquierdo, A.; Ranasinghe, T.; Mikaberidze, B.; Ostermann, S.; Gurgurov, D.; Mueller, P.; Borg, C.; Šimko, M.; and Eggleston, L. E.\n\n\n \n\n\n\n . 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"SIG$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{estevanell-valladaresSIG2025,\n\ttitle = {{SIG}},\n\turl = {https://aclanthology.org/volumes/2025.lowresnlp-1/},\n\turldate = {2026-02-26},\n\tauthor = {Estevanell-Valladares, Ernesto Luis and Picazo-Izquierdo, Alicia and Ranasinghe, Tharindu and Mikaberidze, Besik and Ostermann, Simon and Gurgurov, Daniil and Mueller, Philipp and Borg, Claudia and Šimko, Marián and Eggleston, Liam Enzo},\n\tyear = {2025},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Semeval-2025 task 7: Multilingual and crosslingual fact-checked claim retrieval.\n \n \n \n \n\n\n \n Peng, Q.; Moro, R.; Gregor, M.; Srba, I.; Ostermann, S.; Šimko, M.; Podroužek, J.; Mesarčík, M.; Kopčan, J.; and Søgaard, A.\n\n\n \n\n\n\n In Proceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025), pages 2498–2511, 2025. \n \n\n\n\n
\n\n\n\n \n \n $\"Semeval-2025$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{pengSemeval2025Task72025,\n\ttitle = {Semeval-2025 task 7: {Multilingual} and crosslingual fact-checked claim retrieval},\n\tshorttitle = {Semeval-2025 task 7},\n\turl = {https://aclanthology.org/2025.semeval-1.323/},\n\turldate = {2026-02-26},\n\tbooktitle = {Proceedings of the 19th {International} {Workshop} on {Semantic} {Evaluation} ({SemEval}-2025)},\n\tauthor = {Peng, Qiwei and Moro, Robert and Gregor, Michal and Srba, Ivan and Ostermann, Simon and Šimko, Marián and Podroužek, Juraj and Mesarčík, Matúš and Kopčan, Jaroslav and Søgaard, Anders},\n\tyear = {2025},\n\tpages = {2498--2511},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Saarland-Groningen at NADI 2025 Shared Task: Effective Dialectal Arabic Speech Processing under Data Constraints.\n \n \n \n \n\n\n \n M. Abdullah, B.; Al Ghussin, Y.; Al-Khalili, Z.; Tarik Özyilmaz, Ö.; Valdenegro-Toro, M.; Ostermann, S.; and Klakow, D.\n\n\n \n\n\n\n In Darwish, K.; Ali, A.; Abu Farha, I.; Touileb, S.; Zitouni, I.; Abdelali, A.; Al-Ghamdi, S.; Alkhereyf, S.; Zaghouani, W.; Khalifa, S.; AlKhamissi, B.; Almatham, R.; Hamed, I.; Alyafeai, Z.; Alowisheq, A.; Inoue, G.; Mrini, K.; and Alshammari, W., editor(s), Proceedings of The Third Arabic Natural Language Processing Conference: Shared Tasks, pages 745–751, Suzhou, China, November 2025. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"Saarland-Groningen$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{m.abdullahSaarlandGroningenNADI20252025,\n\taddress = {Suzhou, China},\n\ttitle = {Saarland-{Groningen} at {NADI} 2025 {Shared} {Task}: {Effective} {Dialectal} {Arabic} {Speech} {Processing} under {Data} {Constraints}},\n\tisbn = {979-8-89176-356-2},\n\tshorttitle = {Saarland-{Groningen} at {NADI} 2025 {Shared} {Task}},\n\turl = {https://aclanthology.org/2025.arabicnlp-sharedtasks.102/},\n\tdoi = {10.18653/v1/2025.arabicnlp-sharedtasks.102},\n\turldate = {2026-02-26},\n\tbooktitle = {Proceedings of {The} {Third} {Arabic} {Natural} {Language} {Processing} {Conference}: {Shared} {Tasks}},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {M. Abdullah, Badr and Al Ghussin, Yusser and Al-Khalili, Zena and Tarik Özyilmaz, Ömer and Valdenegro-Toro, Matias and Ostermann, Simon and Klakow, Dietrich},\n\teditor = {Darwish, Kareem and Ali, Ahmed and Abu Farha, Ibrahim and Touileb, Samia and Zitouni, Imed and Abdelali, Ahmed and Al-Ghamdi, Sharefah and Alkhereyf, Sakhar and Zaghouani, Wajdi and Khalifa, Salam and AlKhamissi, Badr and Almatham, Rawan and Hamed, Injy and Alyafeai, Zaid and Alowisheq, Areeb and Inoue, Go and Mrini, Khalil and Alshammari, Waad},\n\tmonth = nov,\n\tyear = {2025},\n\tpages = {745--751},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n PyDSMC: Statistical Model Checking for Neural Agents Using the Gymnasium.\n \n \n \n \n\n\n \n Gros, T. P.; Hartmanns, A.; Hoese, I.; Meyer, J.; Müller, N. J.; and Wolf, V.\n\n\n \n\n\n\n In Quantitative Evaluation of Systems and Formal Modeling and Analysis of Timed Systems: Second International Joint Conference, QEST+ FORMATS 2025, Aarhus, Denmark, August 26–28, 2025, Proceedings, pages 134, 2025. Springer Nature\n \n\n\n\n
\n\n\n\n \n \n $\"PyDSMC:$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{grosPyDSMCStatisticalModel2025,\n\ttitle = {{PyDSMC}: {Statistical} {Model} {Checking} for {Neural} {Agents} {Using} the {Gymnasium}},\n\tshorttitle = {{PyDSMC}},\n\turl = {https://books.google.com/books?hl=en&lr=&id=hI6LEQAAQBAJ&oi=fnd&pg=PA134&dq=info:4qbJJCHbWXQJ:scholar.google.com&ots=EKzaqLcCkN&sig=d8DLFgxILJ_KChKaQyAf-Vs3GMY},\n\turldate = {2026-02-26},\n\tbooktitle = {Quantitative {Evaluation} of {Systems} and {Formal} {Modeling} and {Analysis} of {Timed} {Systems}: {Second} {International} {Joint} {Conference}, {QEST}+ {FORMATS} 2025, {Aarhus}, {Denmark}, {August} 26–28, 2025, {Proceedings}},\n\tpublisher = {Springer Nature},\n\tauthor = {Gros, Timo P. and Hartmanns, Arnd and Hoese, Ivo and Meyer, Joshua and Müller, Nicola J. and Wolf, Verena},\n\tyear = {2025},\n\tpages = {134},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n What language (s) does aya-23 think in? how multilinguality affects internal language representations.\n \n \n \n \n\n\n \n Trinley, K. A.; Nakai, T.; Anikina, T.; and Baeumel, T.\n\n\n \n\n\n\n In Proceedings of the Workshop on Beyond English: Natural Language Processing for all Languages in an Era of Large Language Models, pages 159–171, 2025. \n \n\n\n\n
\n\n\n\n \n \n $\"What$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{trinleyWhatLanguageDoes2025,\n\ttitle = {What language (s) does aya-23 think in? how multilinguality affects internal language representations},\n\tshorttitle = {What language (s) does aya-23 think in?},\n\turl = {https://aclanthology.org/2025.globalnlp-1.18/},\n\turldate = {2026-02-26},\n\tbooktitle = {Proceedings of the {Workshop} on {Beyond} {English}: {Natural} {Language} {Processing} for all {Languages} in an {Era} of {Large} {Language} {Models}},\n\tauthor = {Trinley, Katharina ATT and Nakai, Toshiki and Anikina, Tatiana and Baeumel, Tanja},\n\tyear = {2025},\n\tpages = {159--171},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Visual analysis of action policy behavior: a case study in grid-world driving.\n \n \n \n \n\n\n \n Gros, T. P.; Groß, D.; Kamp, J.; Gumhold, S.; and Hoffman, J.\n\n\n \n\n\n\n In World Conference on Explainable Artificial Intelligence. Springer, Heidelberg, 2025. \n \n\n\n\n
\n\n\n\n \n \n $\"Visual$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{grosVisualAnalysisAction2025,\n\ttitle = {Visual analysis of action policy behavior: a case study in grid-world driving},\n\tshorttitle = {Visual analysis of action policy behavior},\n\turl = {https://mosi.uni-saarland.de/assets/papers/XAI25.pdf},\n\turldate = {2026-02-26},\n\tbooktitle = {World {Conference} on {Explainable} {Artificial} {Intelligence}. {Springer}, {Heidelberg}},\n\tauthor = {Gros, Timo P. and Groß, David and Kamp, Julius and Gumhold, Stefan and Hoffman, J.},\n\tyear = {2025},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n SLR: An Automated Synthesis Framework for Scalable Logical Reasoning.\n \n \n \n \n\n\n \n Helff, L.; Omar, A.; Friedrich, F.; Stammer, W.; Wüst, A.; Woydt, T.; Mitchell, R.; Schramowski, P.; and Kersting, K.\n\n\n \n\n\n\n arXiv e-prints,arXiv–2506. 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"SLR:$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{helffSLRAutomatedSynthesis2025,\n\ttitle = {{SLR}: {An} {Automated} {Synthesis} {Framework} for {Scalable} {Logical} {Reasoning}},\n\tshorttitle = {{SLR}},\n\turl = {https://ui.adsabs.harvard.edu/abs/2025arXiv250615787H/abstract},\n\turldate = {2026-02-26},\n\tjournal = {arXiv e-prints},\n\tauthor = {Helff, Lukas and Omar, Ahmad and Friedrich, Felix and Stammer, Wolfgang and Wüst, Antonia and Woydt, Tim and Mitchell, Rupert and Schramowski, Patrick and Kersting, Kristian},\n\tyear = {2025},\n\tpages = {arXiv--2506},\n}\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Language Arithmetics: Towards Systematic Language Neuron Identification and Manipulation.\n \n \n \n \n\n\n \n Gurgurov, D.; Trinley, K.; Al Ghussin, Y.; Baeumel, T.; Genabith, J. V.; and Ostermann, S.\n\n\n \n\n\n\n In Inui, K.; Sakti, S.; Wang, H.; Wong, D. F.; Bhattacharyya, P.; Banerjee, B.; Ekbal, A.; Chakraborty, T.; and Singh, D. P., editor(s), Proceedings of the 14th International Joint Conference on Natural Language Processing and the 4th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics, pages 2911–2937, Mumbai, India, December 2025. The Asian Federation of Natural Language Processing and The Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"Language$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{gurgurovLanguageArithmeticsSystematic2025,\n\taddress = {Mumbai, India},\n\ttitle = {Language {Arithmetics}: {Towards} {Systematic} {Language} {Neuron} {Identification} and {Manipulation}},\n\tisbn = {979-8-89176-298-5},\n\tshorttitle = {Language {Arithmetics}},\n\turl = {https://aclanthology.org/2025.ijcnlp-long.156/},\n\tabstract = {Large language models (LLMs) exhibit strong multilingual abilities, yet the neural mechanisms behind language-specific processing remain unclear. We analyze language-specific neurons in Llama-3.1-8B, Mistral-Nemo-12B, and Aya-Expanse-8B \\& 32B across 21 typologically diverse languages, identifying neurons that control language behavior. Using the Language Activation Probability Entropy (LAPE) method, we show that these neurons cluster in deeper layers, with non-Latin scripts showing greater specialization. Related languages share overlapping neurons, reflecting internal representations of linguistic proximity.Through language arithmetics, i.e. systematic activation addition and multiplication, we steer models to deactivate unwanted languages and activate desired ones, outperforming established replacement approaches. These interventions effectively guide behavior across five multilingual tasks: language forcing, translation, QA, comprehension, and NLI. Manipulation is more successful for high-resource languages, while typological similarity improves effectiveness. We also demonstrate that neuron steering enhances downstream performance and reveal internal "fallback" mechanisms for language selection when neurons are progressively deactivated. Our code is made publicly available at https://github.com/d-gurgurov/Language-Neurons-Manipulation.},\n\turldate = {2026-01-28},\n\tbooktitle = {Proceedings of the 14th {International} {Joint} {Conference} on {Natural} {Language} {Processing} and the 4th {Conference} of the {Asia}-{Pacific} {Chapter} of the {Association} for {Computational} {Linguistics}},\n\tpublisher = {The Asian Federation of Natural Language Processing and The Association for Computational Linguistics},\n\tauthor = {Gurgurov, Daniil and Trinley, Katharina and Al Ghussin, Yusser and Baeumel, Tanja and Genabith, Josef Van and Ostermann, Simon},\n\teditor = {Inui, Kentaro and Sakti, Sakriani and Wang, Haofen and Wong, Derek F. and Bhattacharyya, Pushpak and Banerjee, Biplab and Ekbal, Asif and Chakraborty, Tanmoy and Singh, Dhirendra Pratap},\n\tmonth = dec,\n\tyear = {2025},\n\tpages = {2911--2937},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Causal discovery on vector-valued variables and consistency-guided aggregation.\n \n \n \n \n\n\n \n Ninad, U.; Wahl, J.; Gerhardus, A.; and Runge, J.\n\n\n \n\n\n\n May 2025.\n arXiv:2505.10476 [stat]\n\n\n\n
\n\n\n\n \n \n $\"Causal$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@misc{ninadCausalDiscoveryVectorvalued2025,\n\ttitle = {Causal discovery on vector-valued variables and consistency-guided aggregation},\n\turl = {http://arxiv.org/abs/2505.10476},\n\tdoi = {10.48550/arXiv.2505.10476},\n\tabstract = {Causal discovery (CD) aims to discover the causal graph underlying the data generation mechanism of observed variables. In many real-world applications, the observed variables are vector-valued, such as in climate science where variables are defined over a spatial grid and the task is called spatio-temporal causal discovery. We motivate CD in vector-valued variable setting while considering different possibilities for the underlying model, and highlight the pitfalls of commonly-used approaches when compared to a fully vectorized approach. Furthermore, often the vector-valued variables are high-dimensional, and aggregations of the variables, such as averages, are considered in interest of efficiency and robustness. In the absence of interventional data, testing for the soundness of aggregate variables as consistent abstractions that map a low-level to a high-level structural causal model (SCM) is hard, and recent works have illustrated the stringency of conditions required for testing consistency. In this work, we take a careful look at the task of vector-valued CD via constraint-based methods, focusing on the problem of consistency of aggregation for this task. We derive three aggregation consistency scores, based on compatibility of independence models and (partial) aggregation, that quantify different aspects of the soundness of an aggregation map for the CD problem. We present the argument that the consistency of causal abstractions must be separated from the task-dependent consistency of aggregation maps. As an actionable conclusion of our findings, we propose a wrapper Adag to optimize a chosen aggregation consistency score for aggregate-CD, to make the output of CD over aggregate variables more reliable. We supplement all our findings with experimental evaluations on synthetic non-time series and spatio-temporal data.},\n\turldate = {2026-01-23},\n\tpublisher = {arXiv},\n\tauthor = {Ninad, Urmi and Wahl, Jonas and Gerhardus, Andreas and Runge, Jakob},\n\tmonth = may,\n\tyear = {2025},\n\tnote = {arXiv:2505.10476 [stat]},\n\tkeywords = {Statistics - Methodology},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Assessing Web Search Credibility and Response Groundedness in Chat Assistants.\n \n \n \n \n\n\n \n Vykopal, I.; Pikuliak, M.; Ostermann, S.; and Šimko, M.\n\n\n \n\n\n\n October 2025.\n arXiv:2510.13749 [cs]\n\n\n\n
\n\n\n\n \n \n $\"Assessing$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@misc{vykopalAssessingWebSearch2025,\n\ttitle = {Assessing {Web} {Search} {Credibility} and {Response} {Groundedness} in {Chat} {Assistants}},\n\turl = {http://arxiv.org/abs/2510.13749},\n\tdoi = {10.48550/arXiv.2510.13749},\n\tabstract = {Chat assistants increasingly integrate web search functionality, enabling them to retrieve and cite external sources. While this promises more reliable answers, it also raises the risk of amplifying misinformation from low-credibility sources. In this paper, we introduce a novel methodology for evaluating assistants' web search behavior, focusing on source credibility and the groundedness of responses with respect to cited sources. Using 100 claims across five misinformation-prone topics, we assess GPT-4o, GPT-5, Perplexity, and Qwen Chat. Our findings reveal differences between the assistants, with Perplexity achieving the highest source credibility, whereas GPT-4o exhibits elevated citation of non-credibility sources on sensitive topics. This work provides the first systematic comparison of commonly used chat assistants for fact-checking behavior, offering a foundation for evaluating AI systems in high-stakes information environments.},\n\turldate = {2026-01-23},\n\tpublisher = {arXiv},\n\tauthor = {Vykopal, Ivan and Pikuliak, Matúš and Ostermann, Simon and Šimko, Marián},\n\tmonth = oct,\n\tyear = {2025},\n\tnote = {arXiv:2510.13749 [cs]},\n\tkeywords = {Computer Science - Computation and Language},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n The AI Language Proficiency Monitor – Tracking the Progress of LLMs on Multilingual Benchmarks.\n \n \n \n \n\n\n \n Pomerenke, D.; Nothnagel, J.; and Ostermann, S.\n\n\n \n\n\n\n July 2025.\n arXiv:2507.08538 [cs]\n\n\n\n
\n\n\n\n \n \n $\"The$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n \n \n 1 download\n \n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@misc{pomerenkeAILanguageProficiency2025,\n\ttitle = {The {AI} {Language} {Proficiency} {Monitor} -- {Tracking} the {Progress} of {LLMs} on {Multilingual} {Benchmarks}},\n\turl = {http://arxiv.org/abs/2507.08538},\n\tdoi = {10.48550/arXiv.2507.08538},\n\tabstract = {To ensure equitable access to the benefits of large language models (LLMs), it is essential to evaluate their capabilities across the world's languages. We introduce the AI Language Proficiency Monitor, a comprehensive multilingual benchmark that systematically assesses LLM performance across up to 200 languages, with a particular focus on low-resource languages. Our benchmark aggregates diverse tasks including translation, question answering, math, and reasoning, using datasets such as FLORES+, MMLU, GSM8K, TruthfulQA, and ARC. We provide an open-source, auto-updating leaderboard and dashboard that supports researchers, developers, and policymakers in identifying strengths and gaps in model performance. In addition to ranking models, the platform offers descriptive insights such as a global proficiency map and trends over time. By complementing and extending prior multilingual benchmarks, our work aims to foster transparency, inclusivity, and progress in multilingual AI. The system is available at https://huggingface.co/spaces/fair-forward/evals-for-every-language.},\n\turldate = {2026-01-23},\n\tpublisher = {arXiv},\n\tauthor = {Pomerenke, David and Nothnagel, Jonas and Ostermann, Simon},\n\tmonth = jul,\n\tyear = {2025},\n\tnote = {arXiv:2507.08538 [cs]},\n\tkeywords = {Computer Science - Computation and Language},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Sparse Subnetwork Enhancement for Underrepresented Languages in Large Language Models.\n \n \n \n \n\n\n \n Gurgurov, D.; Genabith, J. v.; and Ostermann, S.\n\n\n \n\n\n\n October 2025.\n arXiv:2510.13580 [cs]\n\n\n\n
\n\n\n\n \n \n $\"Sparse$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@misc{gurgurovSparseSubnetworkEnhancement2025,\n\ttitle = {Sparse {Subnetwork} {Enhancement} for {Underrepresented} {Languages} in {Large} {Language} {Models}},\n\turl = {http://arxiv.org/abs/2510.13580},\n\tdoi = {10.48550/arXiv.2510.13580},\n\tabstract = {Large language models exhibit uneven performance across languages, with substantial gaps between high- and low-resource languages. We present a framework for enhancing monolingual capabilities of LLMs in underrepresented languages while preserving their general-purpose performance through targeted fine-tuning of language-specific subnetworks. Our approach identifies language-specific neurons using Language Activation Probability Entropy and fine-tunes only the weights associated with these neurons, a dedicated subnetwork, on target-language data. Experiments on Llama-3.1-8B and Mistral-Nemo-12B across 12 mid- and low-resource languages demonstrate that our method consistently outperforms full fine-tuning, FFN-only fine-tuning, LoRA adaptation, and random subset fine-tuning baselines while efficiently updating only up to 1\\% of model parameters. Beyond performance improvements, we observe enhanced favorable training dynamics, cross-lingual representational alignment, and systematic weight update changes. To facilitate future research, we release language-specific neuron identifications for over 100 languages as well as our adaptation pipeline, offering a cost-effective pathway for adapting state-of-the-art models to underrepresented languages.},\n\turldate = {2026-01-23},\n\tpublisher = {arXiv},\n\tauthor = {Gurgurov, Daniil and Genabith, Josef van and Ostermann, Simon},\n\tmonth = oct,\n\tyear = {2025},\n\tnote = {arXiv:2510.13580 [cs]},\n\tkeywords = {Computer Science - Computation and Language},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Small Models, Big Impact: Efficient Corpus and Graph-Based Adaptation of Small Multilingual Language Models for Low-Resource Languages.\n \n \n \n \n\n\n \n Gurgurov, D.; Vykopal, I.; Genabith, J. v.; and Ostermann, S.\n\n\n \n\n\n\n February 2025.\n arXiv:2502.10140 [cs]\n\n\n\n
\n\n\n\n \n \n $\"Small$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@misc{gurgurovSmallModelsBig2025,\n\ttitle = {Small {Models}, {Big} {Impact}: {Efficient} {Corpus} and {Graph}-{Based} {Adaptation} of {Small} {Multilingual} {Language} {Models} for {Low}-{Resource} {Languages}},\n\tshorttitle = {Small {Models}, {Big} {Impact}},\n\turl = {http://arxiv.org/abs/2502.10140},\n\tdoi = {10.48550/arXiv.2502.10140},\n\tabstract = {Low-resource languages (LRLs) face significant challenges in natural language processing (NLP) due to limited data. While current state-of-the-art large language models (LLMs) still struggle with LRLs, smaller multilingual models (mLMs) such as mBERT and XLM-R offer greater promise due to a better fit of their capacity to low training data sizes. This study systematically investigates parameter-efficient adapter-based methods for adapting mLMs to LRLs, evaluating three architectures: Sequential Bottleneck, Invertible Bottleneck, and Low-Rank Adaptation. Using unstructured text from GlotCC and structured knowledge from ConceptNet, we show that small adaptation datasets (e.g., up to 1 GB of free-text or a few MB of knowledge graph data) yield gains in intrinsic (masked language modeling) and extrinsic tasks (topic classification, sentiment analysis, and named entity recognition). We find that Sequential Bottleneck adapters excel in language modeling, while Invertible Bottleneck adapters slightly outperform other methods on downstream tasks due to better embedding alignment and larger parameter counts. Adapter-based methods match or outperform full fine-tuning while using far fewer parameters, and smaller mLMs prove more effective for LRLs than massive LLMs like LLaMA-3, GPT-4, and DeepSeek-R1-based distilled models. While adaptation improves performance, pre-training data size remains the dominant factor, especially for languages with extensive pre-training coverage.},\n\turldate = {2026-01-23},\n\tpublisher = {arXiv},\n\tauthor = {Gurgurov, Daniil and Vykopal, Ivan and Genabith, Josef van and Ostermann, Simon},\n\tmonth = feb,\n\tyear = {2025},\n\tnote = {arXiv:2502.10140 [cs]},\n\tkeywords = {Computer Science - Computation and Language},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n OpenFActScore: Open-Source Atomic Evaluation of Factuality in Text Generation.\n \n \n \n \n\n\n \n Lage, L. F.; and Ostermann, S.\n\n\n \n\n\n\n July 2025.\n arXiv:2507.05965 [cs]\n\n\n\n
\n\n\n\n \n \n $\"OpenFActScore:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n\n\n\n

@misc{lageOpenFActScoreOpenSourceAtomic2025,\n\ttitle = {{OpenFActScore}: {Open}-{Source} {Atomic} {Evaluation} of {Factuality} in {Text} {Generation}},\n\tshorttitle = {{OpenFActScore}},\n\turl = {http://arxiv.org/abs/2507.05965},\n\tdoi = {10.48550/arXiv.2507.05965},\n\tabstract = {We introduce OpenFActScore, an open-source implementation of the FActScore framework for evaluating the factuality of text generated by large language models (LLMs). FActScore evaluates the factual accuracy of long-form text by using Atomic Fact Generation (AFG) to extract individual factual claims and Atomic Fact Validation (AFV) to verify each claim against a trusted knowledge source. While the original FActScore relies on closed-source and commercial models such as InstructGPT and ChatGPT, OpenFActScore enables the use of any Hugging Face-compatible model for both AFG and AFV. We provide a detailed technical overview of our implementation, highlighting design choices and modifications made to support open models. We evaluate multiple open-source LLMs on both AFG and AFV using the original FActScore benchmark, reporting BERTScore-F1 for AFG and Error Rate relative to human annotations for AFV. Our results show that open models can approximate the performance of closed-source systems, with Gemma achieving the best overall performance, and our final setup obtains a 0.99 Pearson correlation with the original FActScore experiments. OpenFActScore promotes transparency, reproducibility, and cost-effective evaluation, and is available at: https://github.com/lflage/OpenFActScore.},\n\turldate = {2026-01-23},\n\tpublisher = {arXiv},\n\tauthor = {Lage, Lucas Fonseca and Ostermann, Simon},\n\tmonth = jul,\n\tyear = {2025},\n\tnote = {arXiv:2507.05965 [cs]},\n\tkeywords = {Computer Science - Artificial Intelligence, Computer Science - Computation and Language},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n On Multilingual Encoder Language Model Compression for Low-Resource Languages.\n \n \n \n \n\n\n \n Gurgurov, D.; Gregor, M.; Genabith, J. v.; and Ostermann, S.\n\n\n \n\n\n\n November 2025.\n arXiv:2505.16956 [cs]\n\n\n\n
\n\n\n\n \n \n $\"On$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@misc{gurgurovMultilingualEncoderLanguage2025,\n\ttitle = {On {Multilingual} {Encoder} {Language} {Model} {Compression} for {Low}-{Resource} {Languages}},\n\turl = {http://arxiv.org/abs/2505.16956},\n\tdoi = {10.48550/arXiv.2505.16956},\n\tabstract = {In this paper, we combine two-step knowledge distillation, structured pruning, truncation, and vocabulary trimming for extremely compressing multilingual encoder-only language models for low-resource languages. Our novel approach systematically combines existing techniques and takes them to the extreme, reducing layer depth, feed-forward hidden size, and intermediate layer embedding size to create significantly smaller monolingual models while retaining essential language-specific knowledge. We achieve compression rates of up to 92\\% while maintaining competitive performance, with average drops of 2-10\\% for moderate compression and 8-13\\% at maximum compression in four downstream tasks, including sentiment analysis, topic classification, named entity recognition, and part-of-speech tagging, across three low-resource languages. Notably, the performance degradation correlates with the amount of language-specific data in the teacher model, with larger datasets resulting in smaller performance losses. Additionally, we conduct ablation studies to identify the best practices for multilingual model compression using these techniques.},\n\turldate = {2026-01-23},\n\tpublisher = {arXiv},\n\tauthor = {Gurgurov, Daniil and Gregor, Michal and Genabith, Josef van and Ostermann, Simon},\n\tmonth = nov,\n\tyear = {2025},\n\tnote = {arXiv:2505.16956 [cs]},\n\tkeywords = {Computer Science - Computation and Language},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Internal Incoherency Scores for Constraint-based Causal Discovery Algorithms.\n \n \n \n \n\n\n \n Faltenbacher, S.; Wahl, J.; Herman, R.; and Runge, J.\n\n\n \n\n\n\n February 2025.\n arXiv:2502.14719 [stat]\n\n\n\n
\n\n\n\n \n \n $\"Internal$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n\n\n\n

@misc{faltenbacherInternalIncoherencyScores2025,\n\ttitle = {Internal {Incoherency} {Scores} for {Constraint}-based {Causal} {Discovery} {Algorithms}},\n\turl = {http://arxiv.org/abs/2502.14719},\n\tdoi = {10.48550/arXiv.2502.14719},\n\tabstract = {Causal discovery aims to infer causal graphs from observational or experimental data. Methods such as the popular PC algorithm are based on conditional independence testing and utilize enabling assumptions, such as the faithfulness assumption, for their inferences. In practice, these assumptions, as well as the functional assumptions inherited from the chosen conditional independence test, are typically taken as a given and not further tested for their validity on the data. In this work, we propose internal coherency scores that allow testing for assumption violations and finite sample errors, whenever detectable without requiring ground truth or further statistical tests. We provide a complete classification of erroneous results, including a distinction between detectable and undetectable errors, and prove that the detectable erroneous results can be measured by our scores. We illustrate our coherency scores on the PC algorithm with simulated and real-world datasets, and envision that testing for internal coherency can become a standard tool in applying constraint-based methods, much like a suite of tests is used to validate the assumptions of classical regression analysis.},\n\turldate = {2026-01-23},\n\tpublisher = {arXiv},\n\tauthor = {Faltenbacher, Sofia and Wahl, Jonas and Herman, Rebecca and Runge, Jakob},\n\tmonth = feb,\n\tyear = {2025},\n\tnote = {arXiv:2502.14719 [stat]},\n\tkeywords = {Computer Science - Machine Learning, Statistics - Machine Learning},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Beyond Transparency: Evaluating Explainability in AI-Supported Fact-Checking.\n \n \n \n \n\n\n \n Schmitt, V.; Bezzaoui, I.; Jakob, C.; Sahitaj, P.; Wang, Q.; Hilbert, A.; Upravitelev, M.; Fegert, J.; Möller, S.; and Solopova, V.\n\n\n \n\n\n\n In Proceedings of the 4th ACM International Workshop on Multimedia AI against Disinformation, pages 63–72, Chicago USA, June 2025. ACM\n \n\n\n\n
\n\n\n\n \n \n $\"Beyond$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{schmittTransparencyEvaluatingExplainability2025,\n\taddress = {Chicago USA},\n\ttitle = {Beyond {Transparency}: {Evaluating} {Explainability} in {AI}-{Supported} {Fact}-{Checking}},\n\tisbn = {979-8-4007-1891-5},\n\tshorttitle = {Beyond {Transparency}},\n\turl = {https://dl.acm.org/doi/10.1145/3733567.3735566},\n\tdoi = {10.1145/3733567.3735566},\n\tlanguage = {en},\n\turldate = {2026-01-23},\n\tbooktitle = {Proceedings of the 4th {ACM} {International} {Workshop} on {Multimedia} {AI} against {Disinformation}},\n\tpublisher = {ACM},\n\tauthor = {Schmitt, Vera and Bezzaoui, Isabel and Jakob, Charlott and Sahitaj, Premtim and Wang, Qianli and Hilbert, Arthur and Upravitelev, Max and Fegert, Jonas and Möller, Sebastian and Solopova, Veronika},\n\tmonth = jun,\n\tyear = {2025},\n\tpages = {63--72},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Secure Human Oversight of AI: Exploring the Attack Surface of Human Oversight.\n \n \n \n \n\n\n \n Ditz, J. C.; Lazar, V.; Lichtmeß, E.; Plesch, C.; Heck, M.; Baum, K.; and Langer, M.\n\n\n \n\n\n\n September 2025.\n arXiv:2509.12290 [cs]\n\n\n\n
\n\n\n\n \n \n $\"Secure$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n\n\n\n

@misc{ditzSecureHumanOversight2025,\n\ttitle = {Secure {Human} {Oversight} of {AI}: {Exploring} the {Attack} {Surface} of {Human} {Oversight}},\n\tshorttitle = {Secure {Human} {Oversight} of {AI}},\n\turl = {http://arxiv.org/abs/2509.12290},\n\tdoi = {10.48550/arXiv.2509.12290},\n\tabstract = {Human oversight of AI is promoted as a safeguard against risks such as inaccurate outputs, system malfunctions, or violations of fundamental rights, and is mandated in regulation like the European AI Act. Yet debates on human oversight have largely focused on its effectiveness, while overlooking a critical dimension: the security of human oversight. We argue that human oversight creates a new attack surface within the safety, security, and accountability architecture of AI operations. Drawing on cybersecurity perspectives, we analyze attack vectors that threaten the requirements of effective human oversight, thereby undermining the safety of AI operations. Such attacks may target the AI system, its communication with oversight personnel, or the personnel themselves. We then outline hardening strategies to mitigate these risks. Our contributions are: (1) introducing a security perspective on human oversight, and (2) providing an overview of attack vectors and hardening strategies to enable secure human oversight of AI.},\n\turldate = {2026-01-23},\n\tpublisher = {arXiv},\n\tauthor = {Ditz, Jonas C. and Lazar, Veronika and Lichtmeß, Elmar and Plesch, Carola and Heck, Matthias and Baum, Kevin and Langer, Markus},\n\tmonth = sep,\n\tyear = {2025},\n\tnote = {arXiv:2509.12290 [cs]},\n\tkeywords = {Computer Science - Computers and Society, Computer Science - Cryptography and Security, Computer Science - Human-Computer Interaction},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Large Language Models for Multilingual Previously Fact-Checked Claim Detection.\n \n \n \n \n\n\n \n Vykopal, I.; Pikuliak, M.; Ostermann, S.; Anikina, T.; Gregor, M.; and Šimko, M.\n\n\n \n\n\n\n September 2025.\n arXiv:2503.02737 [cs]\n\n\n\n
\n\n\n\n \n \n $\"Large$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@misc{vykopalLargeLanguageModels2025,\n\ttitle = {Large {Language} {Models} for {Multilingual} {Previously} {Fact}-{Checked} {Claim} {Detection}},\n\turl = {http://arxiv.org/abs/2503.02737},\n\tdoi = {10.48550/arXiv.2503.02737},\n\tabstract = {In our era of widespread false information, human fact-checkers often face the challenge of duplicating efforts when verifying claims that may have already been addressed in other countries or languages. As false information transcends linguistic boundaries, the ability to automatically detect previously fact-checked claims across languages has become an increasingly important task. This paper presents the first comprehensive evaluation of large language models (LLMs) for multilingual previously fact-checked claim detection. We assess seven LLMs across 20 languages in both monolingual and cross-lingual settings. Our results show that while LLMs perform well for high-resource languages, they struggle with low-resource languages. Moreover, translating original texts into English proved to be beneficial for low-resource languages. These findings highlight the potential of LLMs for multilingual previously fact-checked claim detection and provide a foundation for further research on this promising application of LLMs.},\n\turldate = {2026-01-23},\n\tpublisher = {arXiv},\n\tauthor = {Vykopal, Ivan and Pikuliak, Matúš and Ostermann, Simon and Anikina, Tatiana and Gregor, Michal and Šimko, Marián},\n\tmonth = sep,\n\tyear = {2025},\n\tnote = {arXiv:2503.02737 [cs]},\n\tkeywords = {Computer Science - Computation and Language},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Unitless Unrestricted Markov-Consistent SCM Generation: Better Benchmark Datasets for Causal Discovery.\n \n \n \n \n\n\n \n Herman, R. J.; Wahl, J.; Ninad, U.; and Runge, J.\n\n\n \n\n\n\n December 2025.\n arXiv:2503.17037 [cs]\n\n\n\n
\n\n\n\n \n \n $\"Unitless$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@misc{hermanUnitlessUnrestrictedMarkovConsistent2025,\n\ttitle = {Unitless {Unrestricted} {Markov}-{Consistent} {SCM} {Generation}: {Better} {Benchmark} {Datasets} for {Causal} {Discovery}},\n\tshorttitle = {Unitless {Unrestricted} {Markov}-{Consistent} {SCM} {Generation}},\n\turl = {http://arxiv.org/abs/2503.17037},\n\tdoi = {10.48550/arXiv.2503.17037},\n\tabstract = {Causal discovery aims to extract qualitative causal knowledge in the form of causal graphs from data. Because causal ground truth is rarely known in the real world, simulated data plays a vital role in evaluating the performance of the various causal discovery algorithms proposed in the literature. But recent work highlighted certain artifacts of commonly used data generation techniques for a standard class of structural causal models (SCM) that may be nonphysical, including var- and R2-sortability, where the variables' variance and coefficients of determination (R2) after regressing on all other variables, respectively, increase along the causal order. Some causal methods exploit such artifacts, leading to unrealistic expectations for their performance on real-world data. Some modifications have been proposed to remove these artifacts; notably, the internally-standardized structural causal model (iSCM) avoids varsortability and largely alleviates R2-sortability on sparse causal graphs, but exhibits a reversed R2-sortability pattern for denser graphs not featured in their work. We analyze which sortability patterns we expect to see in real data, and propose a method for drawing coefficients that we argue more effectively samples the space of SCMs. Finally, we propose a novel extension of our SCM generation method to the time series setting.},\n\turldate = {2026-01-23},\n\tpublisher = {arXiv},\n\tauthor = {Herman, Rebecca J. and Wahl, Jonas and Ninad, Urmi and Runge, Jakob},\n\tmonth = dec,\n\tyear = {2025},\n\tnote = {arXiv:2503.17037 [cs]},\n\tkeywords = {Computer Science - Machine Learning},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n ‪Trustworthy AI and the EU AI Act: Market Analysis‬.\n \n \n \n \n\n\n \n Meyer-Vitali, A.\n\n\n \n\n\n\n December 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"‪Trustworthy$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@misc{meyer-vitaliTrustworthyAIEU2025,\n\ttitle = {‪{Trustworthy} {AI} and the {EU} {AI} {Act}: {Market} {Analysis}‬},\n\tshorttitle = {‪{Trustworthy} {AI} and the {EU} {AI} {Act}},\n\turl = {https://scholar.google.com/citations?view_op=view_citation&hl=de&user=g13AKugAAAAJ&sortby=pubdate&citation_for_view=g13AKugAAAAJ:isC4tDSrTZIC},\n\tabstract = {‪A Byrne, A Suárez-Cetrulo, R Simón-Carbajo, A Aguirre, A Barnardini, E Lopez Sanchez, A Ramfos, E Tsalapati, C De Majo, D Burema…‬, ‪https://bdva.eu/news/etami-market-analysis/, 2025‬},\n\turldate = {2026-01-23},\n\tauthor = {Meyer-Vitali, André},\n\tmonth = dec,\n\tyear = {2025},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n The Lookahead Limitation: Why Multi-Operand Addition is Hard for LLMs.\n \n \n \n \n\n\n \n Baeumel, T.; Genabith, J. v.; and Ostermann, S.\n\n\n \n\n\n\n February 2025.\n arXiv:2502.19981 [cs]\n\n\n\n
\n\n\n\n \n \n $\"The$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@misc{baeumelLookaheadLimitationWhy2025,\n\ttitle = {The {Lookahead} {Limitation}: {Why} {Multi}-{Operand} {Addition} is {Hard} for {LLMs}},\n\tshorttitle = {The {Lookahead} {Limitation}},\n\turl = {http://arxiv.org/abs/2502.19981},\n\tdoi = {10.48550/arXiv.2502.19981},\n\tabstract = {Autoregressive large language models (LLMs) exhibit impressive performance across various tasks but struggle with simple arithmetic, such as addition of two or more operands. We show that this struggle arises from LLMs' use of a simple one-digit lookahead heuristic, which works fairly well (but not perfect) for two-operand addition but fails in multi-operand cases, where the carry-over logic is more complex. Our probing experiments and digit-wise accuracy evaluation show that LLMs fail precisely where a one-digit lookahead is insufficient to account for cascading carries. We analyze the impact of tokenization strategies on arithmetic performance and show that all investigated models, regardless of tokenization, are inherently limited in the addition of multiple operands due to their reliance on a one-digit lookahead heuristic. Our findings reveal fundamental limitations that prevent LLMs from generalizing to more complex numerical reasoning.},\n\turldate = {2026-01-23},\n\tpublisher = {arXiv},\n\tauthor = {Baeumel, Tanja and Genabith, Josef van and Ostermann, Simon},\n\tmonth = feb,\n\tyear = {2025},\n\tnote = {arXiv:2502.19981 [cs]},\n\tkeywords = {Computer Science - Computation and Language},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Yusser al Ghussin, Josef van Genabith, and Simon Ostermann. 2025. Modular arithmetic: Language models solve math digit by digit.\n \n \n \n \n\n\n \n Baeumel, T.; and Gurgurov, D.\n\n\n \n\n\n\n Preprint. December 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"Yusser$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@article{baeumelYusserGhussinJosef2025,\n\ttitle = {Yusser al {Ghussin}, {Josef} van {Genabith}, and {Simon} {Ostermann}. 2025. {Modular} arithmetic: {Language} models solve math digit by digit},\n\tshorttitle = {Yusser al {Ghussin}, {Josef} van {Genabith}, and {Simon} {Ostermann}. 2025. {Modular} arithmetic},\n\turl = {https://aclanthology.org/2025.findings-ijcnlp.86/},\n\turldate = {2026-01-23},\n\tjournal = {Preprint},\n\tauthor = {Baeumel, Tanja and Gurgurov, Daniil},\n\tmonth = dec,\n\tyear = {2025},\n\tkeywords = {⛔ No DOI found},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n PolBiX: Detecting LLMs' Political Bias in Fact-Checking through X-phemisms.\n \n \n \n \n\n\n \n Jakob, C.; Harbecke, D.; Parschan, P.; Neves, P. W.; and Schmitt, V.\n\n\n \n\n\n\n arXiv preprint arXiv:2509.15335. 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"PolBiX:$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n \n \n 1 download\n \n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@article{jakobPolBiXDetectingLLMs2025,\n\ttitle = {{PolBiX}: {Detecting} {LLMs}' {Political} {Bias} in {Fact}-{Checking} through {X}-phemisms},\n\tshorttitle = {{PolBiX}},\n\turl = {https://aclanthology.org/anthology-files/pdf/findings/2025.findings-emnlp.932.pdf},\n\turldate = {2026-01-23},\n\tjournal = {arXiv preprint arXiv:2509.15335},\n\tauthor = {Jakob, Charlott and Harbecke, David and Parschan, Patrick and Neves, Pia Wenzel and Schmitt, Vera},\n\tyear = {2025},\n\tkeywords = {⛔ No DOI found},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Licence to Scale: A Microservice Simulation Environment for Benchmarking Agentic AI.\n \n \n \n \n\n\n \n Lohse, C.; Selk, A.; Ba, A.; Wahl, J.; and Ruffini, M.\n\n\n \n\n\n\n In Workshop on Scaling Environments for Agents, 2025. \n \n\n\n\n
\n\n\n\n \n \n $\"Licence$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@inproceedings{lohseLicenceScaleMicroservice2025,\n\ttitle = {Licence to {Scale}: {A} {Microservice} {Simulation} {Environment} for {Benchmarking} {Agentic} {AI}},\n\tshorttitle = {Licence to {Scale}},\n\turl = {https://openreview.net/forum?id=03xThlPUxU},\n\turldate = {2026-01-23},\n\tbooktitle = {Workshop on {Scaling} {Environments} for {Agents}},\n\tauthor = {Lohse, Christopher and Selk, Adrian and Ba, Amadou and Wahl, Jonas and Ruffini, Marco},\n\tyear = {2025},\n\tkeywords = {⛔ No DOI found},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Hybrid annotation for propaganda detection: integrating LLM pre-annotations with human intelligence.\n \n \n \n \n\n\n \n Sahitaj, A.; Sahitaj, P.; Solopova, V.; Li, J.; Möller, S.; and Schmitt, V.\n\n\n \n\n\n\n In Proceedings of the Fourth Workshop on NLP for Positive Impact (NLP4PI), pages 215–228, 2025. \n \n\n\n\n
\n\n\n\n \n \n $\"Hybrid$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{sahitajHybridAnnotationPropaganda2025,\n\ttitle = {Hybrid annotation for propaganda detection: integrating {LLM} pre-annotations with human intelligence},\n\tshorttitle = {Hybrid annotation for propaganda detection},\n\turl = {https://aclanthology.org/2025.nlp4pi-1.18/},\n\tdoi = {10.18653/v1/2025.nlp4pi-1.18},\n\turldate = {2026-01-23},\n\tbooktitle = {Proceedings of the {Fourth} {Workshop} on {NLP} for {Positive} {Impact} ({NLP4PI})},\n\tauthor = {Sahitaj, Ariana and Sahitaj, Premtim and Solopova, Veronika and Li, Jiaao and Möller, Sebastian and Schmitt, Vera},\n\tyear = {2025},\n\tpages = {215--228},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Exploring semantic filtering heuristics for efficient claim verification.\n \n \n \n \n\n\n \n Upravitelev, M.; Sahitaj, P.; Hilbert, A.; Solopova, V.; Yang, J.; Feldhus, N.; Anikina, T.; Ostermann, S.; and Schmitt, V.\n\n\n \n\n\n\n In Proceedings of the Eighth Fact Extraction and VERification Workshop (FEVER), pages 229–237, 2025. \n \n\n\n\n
\n\n\n\n \n \n $\"Exploring$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{upravitelevExploringSemanticFiltering2025a,\n\ttitle = {Exploring semantic filtering heuristics for efficient claim verification},\n\turl = {https://aclanthology.org/2025.fever-1.17/},\n\tdoi = {10.18653/v1/2025.fever-1.17},\n\turldate = {2026-01-23},\n\tbooktitle = {Proceedings of the {Eighth} {Fact} {Extraction} and {VERification} {Workshop} ({FEVER})},\n\tauthor = {Upravitelev, Max and Sahitaj, Premtim and Hilbert, Arthur and Solopova, Veronika and Yang, Jing and Feldhus, Nils and Anikina, Tatiana and Ostermann, Simon and Schmitt, Vera},\n\tyear = {2025},\n\tpages = {229--237},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n A rigorous evaluation of llm data generation strategies for low-resource languages.\n \n \n \n \n\n\n \n Anikina, T.; Cegin, J.; Simko, J.; and Ostermann, S.\n\n\n \n\n\n\n In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 8293–8314, 2025. \n \n\n\n\n
\n\n\n\n \n \n $\"A$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{anikinaRigorousEvaluationLlm2025,\n\ttitle = {A rigorous evaluation of llm data generation strategies for low-resource languages},\n\turl = {https://aclanthology.org/2025.emnlp-main.418/},\n\tdoi = {10.18653/v1/2025.emnlp-main.418},\n\turldate = {2026-01-23},\n\tbooktitle = {Proceedings of the 2025 {Conference} on {Empirical} {Methods} in {Natural} {Language} {Processing}},\n\tauthor = {Anikina, Tatiana and Cegin, Jan and Simko, Jakub and Ostermann, Simon},\n\tyear = {2025},\n\tpages = {8293--8314},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Only for the Unseen Languages, Say the Llamas: On the Efficacy of Language Adapters for Cross-lingual Transfer in English-centric LLMs.\n \n \n \n \n\n\n \n Schlenker, J.; Kunz, J.; Anikina, T.; Neumann, G.; and Ostermann, S.\n\n\n \n\n\n\n In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop), pages 849–871, 2025. \n \n\n\n\n
\n\n\n\n \n \n $\"Only$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{schlenkerOnlyUnseenLanguages2025,\n\ttitle = {Only for the {Unseen} {Languages}, {Say} the {Llamas}: {On} the {Efficacy} of {Language} {Adapters} for {Cross}-lingual {Transfer} in {English}-centric {LLMs}},\n\tshorttitle = {Only for the {Unseen} {Languages}, {Say} the {Llamas}},\n\turl = {https://aclanthology.org/2025.acl-srw.62/},\n\tdoi = {10.18653/v1/2025.acl-srw.62},\n\turldate = {2026-01-23},\n\tbooktitle = {Proceedings of the 63rd {Annual} {Meeting} of the {Association} for {Computational} {Linguistics} ({Volume} 4: {Student} {Research} {Workshop})},\n\tauthor = {Schlenker, Julian and Kunz, Jenny and Anikina, Tatiana and Neumann, Günter and Ostermann, Simon},\n\tyear = {2025},\n\tpages = {849--871},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Do LLMs fail in bridging generation?.\n \n \n \n \n\n\n \n Skachkova, N.; Ostermann, S.; van Genabith, J.; and Kiefer, B.\n\n\n \n\n\n\n Journal for Language Technology and Computational Linguistics, 38(2): 77–95. 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"Do$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{skachkovaLLMsFailBridging2025,\n\ttitle = {Do {LLMs} fail in bridging generation?},\n\tvolume = {38},\n\turl = {https://jlcl.org/article/view/286},\n\tdoi = {10.21248/jlcl.38.2025.286},\n\tnumber = {2},\n\turldate = {2026-01-23},\n\tjournal = {Journal for Language Technology and Computational Linguistics},\n\tauthor = {Skachkova, Natalia and Ostermann, Simon and van Genabith, Josef and Kiefer, Bernd},\n\tyear = {2025},\n\tpages = {77--95},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Deepfakes–Unsere neue Realität?.\n \n \n \n \n\n\n \n Schmitt, V.\n\n\n \n\n\n\n VM Verwaltung & Management, 31(5): 259–263. 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"Deepfakes–Unsere$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{schmittDeepfakesUnsereNeue2025,\n\ttitle = {Deepfakes–{Unsere} neue {Realität}?},\n\tvolume = {31},\n\turl = {https://scholar.google.com/scholar?cluster=13564473952997285219&hl=en&oi=scholarr},\n\tdoi = {10.5771/0947-9856-2025-5-259},\n\tnumber = {5},\n\turldate = {2026-01-23},\n\tjournal = {VM Verwaltung \\& Management},\n\tauthor = {Schmitt, Vera},\n\tyear = {2025},\n\tpages = {259--263},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n AutoPsyC: Automatic Recognition of Psychodynamic Conflicts from Semi-structured Interviews with Large Language Models.\n \n \n \n \n\n\n \n Hossain, S.; Ostermann, S.; Gebhard, P.; Benecke, C.; van Genabith, J.; and Müller, P.\n\n\n \n\n\n\n In Proceedings of the 10th Workshop on Computational Linguistics and Clinical Psychology (CLPsych 2025), pages 12–25, 2025. \n \n\n\n\n
\n\n\n\n \n \n $\"AutoPsyC:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{hossainAutoPsyCAutomaticRecognition2025,\n\ttitle = {{AutoPsyC}: {Automatic} {Recognition} of {Psychodynamic} {Conflicts} from {Semi}-structured {Interviews} with {Large} {Language} {Models}},\n\tshorttitle = {{AutoPsyC}},\n\turl = {https://scholar.google.com/scholar?cluster=9409479046806687550&hl=en&oi=scholarr},\n\tdoi = {10.18653/v1/2025.clpsych-1.2},\n\turldate = {2026-01-23},\n\tbooktitle = {Proceedings of the 10th {Workshop} on {Computational} {Linguistics} and {Clinical} {Psychology} ({CLPsych} 2025)},\n\tauthor = {Hossain, Sayed and Ostermann, Simon and Gebhard, Patrick and Benecke, Cord and van Genabith, Josef and Müller, Philipp},\n\tyear = {2025},\n\tpages = {12--25},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n 3. AI Development and Governance: Navigating Trust, Transparency, Innovation, and the Challenges of Information Warfare.\n \n \n \n \n\n\n \n Schmitt, V.\n\n\n \n\n\n\n Information Warfare,50. 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"3.$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@article{schmitt3AIDevelopment2025,\n\ttitle = {3. {AI} {Development} and {Governance}: {Navigating} {Trust}, {Transparency}, {Innovation}, and the {Challenges} of {Information} {Warfare}},\n\tshorttitle = {3. {AI} {Development} and {Governance}},\n\turl = {https://www.isdp.eu/wp-content/uploads/2025/11/Special-Paper-New-Tech-A5-final2.pdf#page=52},\n\turldate = {2026-01-23},\n\tjournal = {Information Warfare},\n\tauthor = {Schmitt, Vera},\n\tyear = {2025},\n\tkeywords = {⛔ No DOI found},\n\tpages = {50},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n DIN SPEC 91527:2025-12, Goals, Methods and Metrics for Automated/Semi-Automated Runtime Monitoring of AI Systems for Non-Adversarial Performance Degradations.\n \n \n \n \n\n\n \n Weimer, D.; Meyer-Vitali, A.; Sielemann, A.; Ziehn, J.; Zhou, J.; Bulut, Y.; Graf, M.; and Krauß, S.\n\n\n \n\n\n\n December 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"DIN$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@misc{weimerDINSPEC915272025,\n\ttitle = {{DIN} {SPEC} 91527:2025-12, {Goals}, {Methods} and {Metrics} for {Automated}/{Semi}-{Automated} {Runtime} {Monitoring} of {AI}  {Systems} for {Non}-{Adversarial} {Performance} {Degradations}},\n\tshorttitle = {{DIN} {SPEC} 91527},\n\turl = {https://www.dinmedia.de/de/-/-/395753303},\n\tdoi = {10.31030/3649777},\n\turldate = {2025-11-21},\n\tpublisher = {DIN Media GmbH},\n\tauthor = {Weimer, Daniel and Meyer-Vitali, André and Sielemann, Anne and Ziehn, Jens and Zhou, Jingxing and Bulut, Yunus and Graf, Michael and Krauß, Sebastian},\n\tmonth = dec,\n\tyear = {2025},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Design Patterns for Large Language Model Based Neuro-Symbolic Systems.\n \n \n \n \n\n\n \n De Boer, M.; Smit, Q.; Van Bekkum, M.; Meyer-Vitali, A.; and Schmid, T.\n\n\n \n\n\n\n Neurosymbolic Artificial Intelligence, 1: 29498732251377499. September 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"Design$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{deboerDesignPatternsLarge2025,\n\ttitle = {Design {Patterns} for {Large} {Language} {Model} {Based} {Neuro}-{Symbolic} {Systems}},\n\tvolume = {1},\n\tissn = {2949-8732, 2949-8732},\n\turl = {https://journals.sagepub.com/doi/10.1177/29498732251377499},\n\tdoi = {10.1177/29498732251377499},\n\tabstract = {Large language models (LLMs) have been a dominating trend in artificial intelligence (AI) in the past years. At the same time, neuro-symbolic systems employing LLMs have also received increasing interest due to their advantages over purely statistical generative models: They can make explicit use of expert knowledge and can be understood and inspected by humans thus providing explainability. However, with an increasing variety of approaches, it is currently difficult to compare the different ways in which designing, training, fine-tuning, and applying such approaches take place. In this work, we use and extend the modular design patterns for hybrid learning and reasoning systems and the Boxology language of van Bekkum et al. for this purpose. These patterns provide a general language to describe, compare, and understand the different architectures and methods used for LLM-based neuro-symbolic systems. The primary goal of this work is to support a better understanding of specific classes of such systems, namely LLM-based models that are used in conjunction with knowledge-based (symbolic) systems. In order to demonstrate the usefulness of this approach, we explore existing LLM-based neuro-symbolic architectures and approaches, as well as use cases for these design patterns.},\n\tlanguage = {en},\n\turldate = {2025-10-02},\n\tjournal = {Neurosymbolic Artificial Intelligence},\n\tauthor = {De Boer, Maaike and Smit, Quirine and Van Bekkum, Michael and Meyer-Vitali, André and Schmid, Thomas},\n\tmonth = sep,\n\tyear = {2025},\n\tpages = {29498732251377499},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Cross-Refine: Improving Natural Language Explanation Generation by Learning in Tandem.\n \n \n \n \n\n\n \n Wang, Q.; Anikina, T.; Feldhus, N.; Ostermann, S.; Möller, S.; and Schmitt, V.\n\n\n \n\n\n\n In Rambow, O.; Wanner, L.; Apidianaki, M.; Al-Khalifa, H.; Eugenio, B. D.; and Schockaert, S., editor(s), Proceedings of the 31st International Conference on Computational Linguistics, pages 1150–1167, Abu Dhabi, UAE, January 2025. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"Cross-Refine:$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{wangCrossRefineImprovingNatural2025,\n\taddress = {Abu Dhabi, UAE},\n\ttitle = {Cross-{Refine}: {Improving} {Natural} {Language} {Explanation} {Generation} by {Learning} in {Tandem}},\n\tshorttitle = {Cross-{Refine}},\n\turl = {https://aclanthology.org/2025.coling-main.77/},\n\tabstract = {Natural language explanations (NLEs) are vital for elucidating the reasoning behind large language model (LLM) decisions. Many techniques have been developed to generate NLEs using LLMs. However, like humans, LLMs might not always produce optimal NLEs on first attempt. Inspired by human learning processes, we introduce Cross-Refine, which employs role modeling by deploying two LLMs as generator and critic, respectively. The generator outputs a first NLE and then refines this initial explanation using feedback and suggestions provided by the critic. Cross-Refine does not require any supervised training data or additional training. We validate Cross-Refine across three NLP tasks using three state-of-the-art open-source LLMs through automatic and human evaluation. We select Self-Refine (Madaan et al., 2023) as the baseline, which only utilizes self-feedback to refine the explanations. Our findings from automatic evaluation and a user study indicate that Cross-Refine outperforms Self-Refine. Meanwhile, Cross-Refine can perform effectively with less powerful LLMs, whereas Self-Refine only yields strong results with ChatGPT. Additionally, we conduct an ablation study to assess the importance of feedback and suggestions. Both of them play an important role in refining explanations. We further evaluate Cross-Refine on a bilingual dataset in English and German.},\n\turldate = {2025-05-22},\n\tbooktitle = {Proceedings of the 31st {International} {Conference} on {Computational} {Linguistics}},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Wang, Qianli and Anikina, Tatiana and Feldhus, Nils and Ostermann, Simon and Möller, Sebastian and Schmitt, Vera},\n\teditor = {Rambow, Owen and Wanner, Leo and Apidianaki, Marianna and Al-Khalifa, Hend and Eugenio, Barbara Di and Schockaert, Steven},\n\tmonth = jan,\n\tyear = {2025},\n\tpages = {1150--1167},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n GrEmLIn: A Repository of Green Baseline Embeddings for 87 Low-Resource Languages Injected with Multilingual Graph Knowledge.\n \n \n \n \n\n\n \n Gurgurov, D.; Kumar, R.; and Ostermann, S.\n\n\n \n\n\n\n In Chiruzzo, L.; Ritter, A.; and Wang, L., editor(s), Findings of the Association for Computational Linguistics: NAACL 2025, pages 1204–1221, Albuquerque, New Mexico, April 2025. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"GrEmLIn:$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{gurgurovGrEmLInRepositoryGreen2025,\n\taddress = {Albuquerque, New Mexico},\n\ttitle = {{GrEmLIn}: {A} {Repository} of {Green} {Baseline} {Embeddings} for 87 {Low}-{Resource} {Languages} {Injected} with {Multilingual} {Graph} {Knowledge}},\n\tisbn = {979-8-89176-195-7},\n\tshorttitle = {{GrEmLIn}},\n\turl = {https://aclanthology.org/2025.findings-naacl.67/},\n\turldate = {2025-05-22},\n\tbooktitle = {Findings of the {Association} for {Computational} {Linguistics}: {NAACL} 2025},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Gurgurov, Daniil and Kumar, Rishu and Ostermann, Simon},\n\teditor = {Chiruzzo, Luis and Ritter, Alan and Wang, Lu},\n\tmonth = apr,\n\tyear = {2025},\n\tpages = {1204--1221},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n FitCF: A Framework for Automatic Feature Importance-guided Counterfactual Example Generation.\n \n \n \n \n\n\n \n Wang, Q.; Feldhus, N.; Ostermann, S.; Villa-Arenas, L. F.; Möller, S.; and Schmitt, V.\n\n\n \n\n\n\n May 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"FitCF:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n\n\n\n

@misc{wangFitCFFrameworkAutomatic2025,\n\ttitle = {{FitCF}: {A} {Framework} for {Automatic} {Feature} {Importance}-guided {Counterfactual} {Example} {Generation}},\n\tshorttitle = {{FitCF}},\n\turl = {http://arxiv.org/abs/2501.00777},\n\tdoi = {10.48550/arXiv.2501.00777},\n\tabstract = {Counterfactual examples are widely used in natural language processing (NLP) as valuable data to improve models, and in explainable artificial intelligence (XAI) to understand model behavior. The automated generation of counterfactual examples remains a challenging task even for large language models (LLMs), despite their impressive performance on many tasks. In this paper, we first introduce ZeroCF, a faithful approach for leveraging important words derived from feature attribution methods to generate counterfactual examples in a zero-shot setting. Second, we present a new framework, FitCF, which further verifies aforementioned counterfactuals by label flip verification and then inserts them as demonstrations for few-shot prompting, outperforming two state-of-the-art baselines. Through ablation studies, we identify the importance of each of FitCF's core components in improving the quality of counterfactuals, as assessed through flip rate, perplexity, and similarity measures. Furthermore, we show the effectiveness of LIME and Integrated Gradients as backbone attribution methods for FitCF and find that the number of demonstrations has the largest effect on performance. Finally, we reveal a strong correlation between the faithfulness of feature attribution scores and the quality of generated counterfactuals.},\n\turldate = {2025-05-22},\n\tpublisher = {arXiv},\n\tauthor = {Wang, Qianli and Feldhus, Nils and Ostermann, Simon and Villa-Arenas, Luis Felipe and Möller, Sebastian and Schmitt, Vera},\n\tmonth = may,\n\tyear = {2025},\n\tkeywords = {Computer Science - Computation and Language, Computer Science - Machine Learning},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Reverse Probing: Evaluating Knowledge Transfer via Finetuned Task Embeddings for Coreference Resolution.\n \n \n \n \n\n\n \n Anikina, T.; Binder, A.; Harbecke, D.; Varanasi, S.; Hennig, L.; Ostermann, S.; Möller, S.; and Genabith, J. V.\n\n\n \n\n\n\n In Adlakha, V.; Chronopoulou, A.; Li, X. L.; Majumder, B. P.; Shi, F.; and Vernikos, G., editor(s), Proceedings of the 10th Workshop on Representation Learning for NLP (RepL4NLP-2025), pages 108–119, Albuquerque, NM, May 2025. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"Reverse$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n \n \n 1 download\n \n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{anikinaReverseProbingEvaluating2025a,\n\taddress = {Albuquerque, NM},\n\ttitle = {Reverse {Probing}: {Evaluating} {Knowledge} {Transfer} via {Finetuned} {Task} {Embeddings} for {Coreference} {Resolution}},\n\tisbn = {979-8-89176-245-9},\n\tshorttitle = {Reverse {Probing}},\n\turl = {https://aclanthology.org/2025.repl4nlp-1.9/},\n\tabstract = {In this work, we reimagine classical probing to evaluate knowledge transfer from simple source to more complex target tasks. Instead of probing frozen representations from a complex source task on diverse simple target probing tasks (as usually done in probing), we explore the effectiveness of embeddings from multiple simple source tasks on a single target task. We select coreference resolution, a linguistically complex problem requiring contextual understanding, as focus target task, and test the usefulness of embeddings from comparably simpler tasks tasks such as paraphrase detection, named entity recognition, and relation extraction. Through systematic experiments, we evaluate the impact of individual and combined task embeddings. Our findings reveal that task embeddings vary significantly in utility for coreference resolution, with semantic similarity tasks (e.g., paraphrase detection) proving most beneficial. Additionally, representations from intermediate layers of fine-tuned models often outperform those from final layers. Combining embeddings from multiple tasks consistently improves performance, with attention-based aggregation yielding substantial gains. These insights shed light on relationships between task-specific representations and their adaptability to complex downstream tasks, encouraging further exploration of embedding-level task transfer. Our source code is publicly available under https://github.com/Cora4NLP/multi-task-knowledge-transfer.},\n\turldate = {2025-05-22},\n\tbooktitle = {Proceedings of the 10th {Workshop} on {Representation} {Learning} for {NLP} ({RepL4NLP}-2025)},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Anikina, Tatiana and Binder, Arne and Harbecke, David and Varanasi, Stalin and Hennig, Leonhard and Ostermann, Simon and Möller, Sebastian and Genabith, Josef Van},\n\teditor = {Adlakha, Vaibhav and Chronopoulou, Alexandra and Li, Xiang Lorraine and Majumder, Bodhisattwa Prasad and Shi, Freda and Vernikos, Giorgos},\n\tmonth = may,\n\tyear = {2025},\n\tpages = {108--119},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Soft Language Prompts for Language Transfer.\n \n \n \n \n\n\n \n Vykopal, I.; Ostermann, S.; and Simko, M.\n\n\n \n\n\n\n In Chiruzzo, L.; Ritter, A.; and Wang, L., editor(s), Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 10294–10313, Albuquerque, New Mexico, April 2025. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"Soft$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n \n \n 3 downloads\n \n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{vykopalSoftLanguagePrompts2025,\n\taddress = {Albuquerque, New Mexico},\n\ttitle = {Soft {Language} {Prompts} for {Language} {Transfer}},\n\tisbn = {979-8-89176-189-6},\n\turl = {https://aclanthology.org/2025.naacl-long.517/},\n\tabstract = {Cross-lingual knowledge transfer, especially between high- and low-resource languages, remains challenging in natural language processing (NLP). This study offers insights for improving cross-lingual NLP applications through the combination of parameter-efficient fine-tuning methods. We systematically explore strategies for enhancing cross-lingual transfer through the incorporation of language-specific and task-specific adapters and soft prompts. We present a detailed investigation of various combinations of these methods, exploring their efficiency across 16 languages, focusing on 10 mid- and low-resource languages. We further present to our knowledge the first use of soft prompts for language transfer, a technique we call soft language prompts. Our findings demonstrate that in contrast to claims of previous work, a combination of language and task adapters does not always work best; instead, combining a soft language prompt with a task adapter outperforms most configurations in many cases.},\n\turldate = {2025-05-22},\n\tbooktitle = {Proceedings of the 2025 {Conference} of the {Nations} of the {Americas} {Chapter} of the {Association} for {Computational} {Linguistics}: {Human} {Language} {Technologies} ({Volume} 1: {Long} {Papers})},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Vykopal, Ivan and Ostermann, Simon and Simko, Marian},\n\teditor = {Chiruzzo, Luis and Ritter, Alan and Wang, Lu},\n\tmonth = apr,\n\tyear = {2025},\n\tpages = {10294--10313},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Developing Effective and Value-Aligned AI Tools for Journalists: 12 Critical Questions to Reflect upon.\n \n \n \n\n\n \n Cordero, J. M.; Henn, T.; Holtel, F.; Sánchez Gómez, J. Á; Arenas, D.; Šipka, A.; and Vollmer, S.\n\n\n \n\n\n\n Journalism Practice,1–20. 2025.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{corderoDevelopingEffectiveValueAligned2025,\n\ttitle = {Developing {Effective} and {Value}-{Aligned} {AI} {Tools} for {Journalists}: 12 {Critical} {Questions} to {Reflect} upon},\n\tjournal = {Journalism Practice},\n\tauthor = {Cordero, José Miguel and Henn, Theresa and Holtel, Frederik and Sánchez Gómez, José Á and Arenas, Diego and Šipka, Andrea and Vollmer, Sebastian},\n\tyear = {2025},\n\tpages = {1--20},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n X Hacking: The Threat of Misguided AutoML.\n \n \n \n\n\n \n Sharma, R.; Redyuk, S.; Mukherjee, S.; Sipka, A.; Vollmer, S.; and Selby, D.\n\n\n \n\n\n\n Accepted at ICML. 2025.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{sharmaHackingThreatMisguided2025,\n\ttitle = {X {Hacking}: {The} {Threat} of {Misguided} {AutoML}},\n\tjournal = {Accepted at ICML},\n\tauthor = {Sharma, Rahul and Redyuk, Sergey and Mukherjee, Sumantrak and Sipka, Andrea and Vollmer, Sebastian and Selby, David},\n\tyear = {2025},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Separation-Based Distance Measures for Causal Graphs.\n \n \n \n \n\n\n \n Wahl, J.; and Runge, J.\n\n\n \n\n\n\n In Li, Y.; Mandt, S.; Agrawal, S.; and Khan, E., editor(s), Proceedings of The 28th International Conference on Artificial Intelligence and Statistics, volume 258, of Proceedings of Machine Learning Research, pages 3412–3420, May 2025. \n \n\n\n\n
\n\n\n\n \n \n $\"Separation-Based$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{wahlSeparationBasedDistanceMeasures2025,\n\tseries = {Proceedings of {Machine} {Learning} {Research}},\n\ttitle = {Separation-{Based} {Distance} {Measures} for {Causal} {Graphs}},\n\tvolume = {258},\n\turl = {https://proceedings.mlr.press/v258/wahl25b.html},\n\tbooktitle = {Proceedings of {The} 28th {International} {Conference} on {Artificial} {Intelligence} and {Statistics}},\n\tauthor = {Wahl, Jonas and Runge, Jakob},\n\teditor = {Li, Yingzhen and Mandt, Stephan and Agrawal, Shipra and Khan, Emtiyaz},\n\tmonth = may,\n\tyear = {2025},\n\tpages = {3412--3420},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Transparent Transparency: Developing a Scheme for Understanding Transparency Requirements.\n \n \n \n\n\n \n Baum, D.; Baum, K.; Zamani, S.; Bennoit, C.; and Werth, D.\n\n\n \n\n\n\n In Steffen, B., editor(s), Bridging the Gap Between AI and Reality, volume 15217, of Lecture Notes in Computer Science (LNCS), pages 55–73, Cham, 2025. Springer Nature Switzerland\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{baumTransparentTransparencyDeveloping2025,\n\taddress = {Cham},\n\tseries = {Lecture {Notes} in {Computer} {Science} ({LNCS})},\n\ttitle = {Transparent {Transparency}: {Developing} a {Scheme} for {Understanding} {Transparency} {Requirements}},\n\tvolume = {15217},\n\tisbn = {978-3-031-75434-0},\n\tdoi = {10.1007/978-3-031-75434-0_5},\n\tbooktitle = {Bridging the {Gap} {Between} {AI} and {Reality}},\n\tpublisher = {Springer Nature Switzerland},\n\tauthor = {Baum, Deborah and Baum, Kevin and Zamani, Sasha and Bennoit, Christian and Werth, Dirk},\n\teditor = {Steffen, Bernhard},\n\tyear = {2025},\n\tpages = {55--73},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Taming the AI Monster: Monitoring of Individual Fairness for Effective Human Oversight.\n \n \n \n \n\n\n \n Baum, K.; Biewer, S.; Hermanns, H.; Hetmank, S.; Langer, M.; Lauber-Rönsberg, A.; and Sterz, S.\n\n\n \n\n\n\n In Neele, T.; and Wijs, A., editor(s), Model Checking Software, volume 14624, of Lecture Notes in Computer Science (LNCS), pages 3–25, Cham, 2025. Springer Nature Switzerland\n \n\n\n\n
\n\n\n\n \n \n $\"Taming$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{baumTamingAIMonster2025,\n\taddress = {Cham},\n\tseries = {Lecture {Notes} in {Computer} {Science} ({LNCS})},\n\ttitle = {Taming the {AI} {Monster}: {Monitoring} of {Individual} {Fairness} for {Effective} {Human} {Oversight}},\n\tvolume = {14624},\n\tisbn = {978-3-031-66149-5},\n\turl = {https://spin-web.github.io/SPIN2024/assets/preproceedings/SPIN2024-paper1.pdf},\n\tdoi = {10.1007/978-3-031-66149-5_1},\n\tbooktitle = {Model {Checking} {Software}},\n\tpublisher = {Springer Nature Switzerland},\n\tauthor = {Baum, Kevin and Biewer, Sebastian and Hermanns, Holger and Hetmank, Sven and Langer, Markus and Lauber-Rönsberg, Anne and Sterz, Sarah},\n\teditor = {Neele, Thomas and Wijs, Anton},\n\tyear = {2025},\n\tpages = {3--25},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Multi-Agent Causal Reinforcement Learning.\n \n \n \n\n\n \n Meyer-Vitali, A.\n\n\n \n\n\n\n In Proceedings of the 13th International Conference on Model-Based Software and Systems Engineering - MBSE-AI Integration, pages 435–442, 2025. SciTePress\n Backup Publisher: INSTICC\n\n\n\n
\n\n\n\n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{meyer-vitaliMultiAgentCausalReinforcement2025,\n\ttitle = {Multi-{Agent} {Causal} {Reinforcement} {Learning}},\n\tisbn = {978-989-758-729-0},\n\tissn = {2184-4348},\n\tdoi = {10.5220/0013400100003896},\n\tbooktitle = {Proceedings of the 13th {International} {Conference} on {Model}-{Based} {Software} and {Systems} {Engineering} - {MBSE}-{AI} {Integration}},\n\tpublisher = {SciTePress},\n\tauthor = {Meyer-Vitali, André},\n\tyear = {2025},\n\tnote = {Backup Publisher: INSTICC},\n\tpages = {435--442},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n RITSA: Toward a Retrieval-Augmented Generation System for Intelligent Transportation Systems Architecture.\n \n \n \n\n\n \n Awadid, A.; Meyer-Vitali, A.; Vereno, D.; and Gagnant, M.\n\n\n \n\n\n\n In Proceedings of the 13th International Conference on Model-Based Software and Systems Engineering - MBSE-AI Integration, pages 466–473, 2025. SciTePress\n Backup Publisher: INSTICC\n\n\n\n
\n\n\n\n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{awadidRITSARetrievalAugmentedGeneration2025,\n\ttitle = {{RITSA}: {Toward} a {Retrieval}-{Augmented} {Generation} {System} for {Intelligent} {Transportation} {Systems} {Architecture}},\n\tisbn = {978-989-758-729-0},\n\tissn = {2184-4348},\n\tdoi = {10.5220/0013443300003896},\n\tbooktitle = {Proceedings of the 13th {International} {Conference} on {Model}-{Based} {Software} and {Systems} {Engineering} - {MBSE}-{AI} {Integration}},\n\tpublisher = {SciTePress},\n\tauthor = {Awadid, Afef and Meyer-Vitali, André and Vereno, Dominik and Gagnant, Maxence},\n\tyear = {2025},\n\tnote = {Backup Publisher: INSTICC},\n\tpages = {466--473},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Safe Reinforcement Learning Through Regret and State Restorations in Evaluation Stages.\n \n \n \n \n\n\n \n Gros, T. P.; Müller, N. J.; Höller, D.; and Wolf, V.\n\n\n \n\n\n\n In Jansen, N.; Junges, S.; Kaminski, B. L.; Matheja, C.; Noll, T.; Quatmann, T.; Stoelinga, M.; and Volk, M., editor(s), Principles of Verification: Cycling the Probabilistic Landscape, volume 15262, pages 18–38. Springer Nature Switzerland, Cham, 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"Safe$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@incollection{grosSafeReinforcementLearning2025,\n\taddress = {Cham},\n\ttitle = {Safe {Reinforcement} {Learning} {Through} {Regret} and {State} {Restorations} in {Evaluation} {Stages}},\n\tvolume = {15262},\n\tisbn = {978-3-031-75777-8 978-3-031-75778-5},\n\turl = {https://link.springer.com/10.1007/978-3-031-75778-5_2},\n\tdoi = {10.1007/978-3-031-75778-5_2},\n\tlanguage = {en},\n\turldate = {2025-05-13},\n\tbooktitle = {Principles of {Verification}: {Cycling} the {Probabilistic} {Landscape}},\n\tpublisher = {Springer Nature Switzerland},\n\tauthor = {Gros, Timo P. and Müller, Nicola J. and Höller, Daniel and Wolf, Verena},\n\teditor = {Jansen, Nils and Junges, Sebastian and Kaminski, Benjamin Lucien and Matheja, Christoph and Noll, Thomas and Quatmann, Tim and Stoelinga, Mariëlle and Volk, Matthias},\n\tyear = {2025},\n\tpages = {18--38},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n How do we assess the trustworthiness of AI? Introducing the trustworthiness assessment model (TrAM).\n \n \n \n \n\n\n \n Schlicker, N.; Baum, K.; Uhde, A.; Sterz, S.; Hirsch, M. C.; and Langer, M.\n\n\n \n\n\n\n Computers in Human Behavior, 170: 108671. September 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"How$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{schlickerHowWeAssess2025,\n\ttitle = {How do we assess the trustworthiness of {AI}? {Introducing} the trustworthiness assessment model ({TrAM})},\n\tvolume = {170},\n\tissn = {07475632},\n\tshorttitle = {How do we assess the trustworthiness of {AI}?},\n\turl = {https://linkinghub.elsevier.com/retrieve/pii/S0747563225001189},\n\tdoi = {10.1016/j.chb.2025.108671},\n\tlanguage = {en},\n\turldate = {2025-05-13},\n\tjournal = {Computers in Human Behavior},\n\tauthor = {Schlicker, Nadine and Baum, Kevin and Uhde, Alarith and Sterz, Sarah and Hirsch, Martin C. and Langer, Markus},\n\tmonth = sep,\n\tyear = {2025},\n\tpages = {108671},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Multilingual Text-to-Image Generation Magnifies Gender Stereotypes and Prompt Engineering May Not Help You.\n \n \n \n \n\n\n \n Friedrich, F.; Hämmerl, K.; Schramowski, P.; Brack, M.; Libovicky, J.; Kersting, K.; and Fraser, A.\n\n\n \n\n\n\n In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics, 2025. arXiv\n \n\n\n\n
\n\n\n\n \n \n $\"Multilingual$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n \n \n\n\n\n

@inproceedings{friedrichMultilingualTexttoImageGeneration2025,\n\ttitle = {Multilingual {Text}-to-{Image} {Generation} {Magnifies} {Gender} {Stereotypes} and {Prompt} {Engineering} {May} {Not} {Help} {You}},\n\tcopyright = {Creative Commons Attribution 4.0 International},\n\turl = {https://arxiv.org/abs/2401.16092},\n\tdoi = {10.48550/ARXIV.2401.16092},\n\tabstract = {Text-to-image generation models have recently achieved astonishing results in image quality, flexibility, and text alignment, and are consequently employed in a fast-growing number of applications. Through improvements in multilingual abilities, a larger community now has access to this technology. However, our results show that multilingual models suffer from significant gender biases just as monolingual models do. Furthermore, the natural expectation that multilingual models will provide similar results across languages does not hold up. Instead, there are important differences between languages. We propose a novel benchmark, MAGBIG, intended to foster research on gender bias in multilingual models. We use MAGBIG to investigate the effect of multilingualism on gender bias in T2I models. To this end, we construct multilingual prompts requesting portraits of people with a certain occupation or trait. Our results show that not only do models exhibit strong gender biases but they also behave differently across languages. Furthermore, we investigate prompt engineering strategies, such as indirect, neutral formulations, to mitigate these biases. Unfortunately, these approaches have limited success and result in worse text-to-image alignment. Consequently, we call for more research into diverse representations across languages in image generators, as well as into steerability to address biased model behavior.},\n\turldate = {2025-05-13},\n\tbooktitle = {Proceedings of the 63rd {Annual} {Meeting} of the {Association} for {Computational} {Linguistics}},\n\tpublisher = {arXiv},\n\tauthor = {Friedrich, Felix and Hämmerl, Katharina and Schramowski, Patrick and Brack, Manuel and Libovicky, Jindrich and Kersting, Kristian and Fraser, Alexander},\n\tyear = {2025},\n\tkeywords = {Computation and Language (cs.CL), Computers and Society (cs.CY), FOS: Computer and information sciences, Machine Learning (cs.LG)},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n LLAVAGUARD: VLM-based Safeguard for Vision Dataset Curation and Safety Assessment.\n \n \n \n\n\n \n Helff, L.; Friedrich, F.; Brack, M.; Schramowski, P.; and Kersting, K.\n\n\n \n\n\n\n In Proceedings of the 41st International Conference on Machine Learning, 2025. \n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{helffLLAVAGUARDVLMbasedSafeguard2025,\n\ttitle = {{LLAVAGUARD}: {VLM}-based {Safeguard} for {Vision} {Dataset} {Curation} and {Safety} {Assessment}},\n\tbooktitle = {Proceedings of the 41st {International} {Conference} on {Machine} {Learning}},\n\tauthor = {Helff, Lukas and Friedrich, Felix and Brack, Manuel and Schramowski, Patrick and Kersting, Kristian},\n\tyear = {2025},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n MSTS: A Multimodal Safety Test Suite for Vision-Language Models.\n \n \n \n \n\n\n \n Röttger, P.; Attanasio, G.; Friedrich, F.; Goldzycher, J.; Parrish, A.; Bhardwaj, R.; Di Bonaventura, C.; Eng, R.; Geagea, G. E. K.; Goswami, S.; Han, J.; Hovy, D.; Jeong, S.; Jeretič, P.; Plaza-del-Arco, F. M.; Rooein, D.; Schramowski, P.; Shaitarova, A.; Shen, X.; Willats, R.; Zugarini, A.; and Vidgen, B.\n\n\n \n\n\n\n 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"MSTS:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n\n\n\n

@misc{rottgerMSTSMultimodalSafety2025,\n\ttitle = {{MSTS}: {A} {Multimodal} {Safety} {Test} {Suite} for {Vision}-{Language} {Models}},\n\tcopyright = {Creative Commons Attribution 4.0 International},\n\tshorttitle = {{MSTS}},\n\turl = {https://arxiv.org/abs/2501.10057},\n\tdoi = {10.48550/ARXIV.2501.10057},\n\tabstract = {Vision-language models (VLMs), which process image and text inputs, are increasingly integrated into chat assistants and other consumer AI applications. Without proper safeguards, however, VLMs may give harmful advice (e.g. how to self-harm) or encourage unsafe behaviours (e.g. to consume drugs). Despite these clear hazards, little work so far has evaluated VLM safety and the novel risks created by multimodal inputs. To address this gap, we introduce MSTS, a Multimodal Safety Test Suite for VLMs. MSTS comprises 400 test prompts across 40 fine-grained hazard categories. Each test prompt consists of a text and an image that only in combination reveal their full unsafe meaning. With MSTS, we find clear safety issues in several open VLMs. We also find some VLMs to be safe by accident, meaning that they are safe because they fail to understand even simple test prompts. We translate MSTS into ten languages, showing non-English prompts to increase the rate of unsafe model responses. We also show models to be safer when tested with text only rather than multimodal prompts. Finally, we explore the automation of VLM safety assessments, finding even the best safety classifiers to be lacking.},\n\turldate = {2025-05-13},\n\tpublisher = {arXiv},\n\tauthor = {Röttger, Paul and Attanasio, Giuseppe and Friedrich, Felix and Goldzycher, Janis and Parrish, Alicia and Bhardwaj, Rishabh and Di Bonaventura, Chiara and Eng, Roman and Geagea, Gaia El Khoury and Goswami, Sujata and Han, Jieun and Hovy, Dirk and Jeong, Seogyeong and Jeretič, Paloma and Plaza-del-Arco, Flor Miriam and Rooein, Donya and Schramowski, Patrick and Shaitarova, Anastassia and Shen, Xudong and Willats, Richard and Zugarini, Andrea and Vidgen, Bertie},\n\tyear = {2025},\n\tkeywords = {Computation and Language (cs.CL), FOS: Computer and information sciences},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n AI Act / KI-VO und Standardisierung.\n \n \n \n \n\n\n \n Meyer-Vitali, A.\n\n\n \n\n\n\n Datenschutz und Datensicherheit - DuD, 49(4): 241–244. April 2025.\n \n\n\n\n
\n\n\n\n \n \n $\"AI$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n \n \n 1 download\n \n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n\n\n\n

@article{meyer-vitaliAIActKIVO2025,\n\ttitle = {{AI} {Act} / {KI}-{VO} und {Standardisierung}},\n\tvolume = {49},\n\tissn = {1862-2607},\n\turl = {https://doi.org/10.1007/s11623-025-2079-2},\n\tdoi = {10.1007/s11623-025-2079-2},\n\tabstract = {Bei der Einführung der KI-Verordnung müssen noch viele Aspekte definiert werden, damit einesorgfältige Prüfung geschehen kann. Wir betrachten die Landschaft der Standardisierung, definierenRisiken und das damit einhergehende Vertrauen. Einige Methoden und Prüfmetriken werden dargestellt},\n\tlanguage = {de},\n\tnumber = {4},\n\turldate = {2025-04-23},\n\tjournal = {Datenschutz und Datensicherheit - DuD},\n\tauthor = {Meyer-Vitali, André},\n\tmonth = apr,\n\tyear = {2025},\n\tkeywords = {Artificial Intelligence},\n\tpages = {241--244},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n

\n \n 2024\n \n \n (37)\n \n \n

\n \n \n

\n \n\n \n \n \n \n \n \n Asymptotic Uncertainty in the Estimation of Frequency Domain Causal Effects for Linear Processes.\n \n \n \n \n\n\n \n Reiter, N.; Wahl, J.; Hegerl, G. C.; and Runge, J.\n\n\n \n\n\n\n June 2024.\n arXiv:2406.18191 [stat]\n\n\n\n
\n\n\n\n \n \n $\"Asymptotic$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n\n\n\n

@misc{reiterAsymptoticUncertaintyEstimation2024,\n\ttitle = {Asymptotic {Uncertainty} in the {Estimation} of {Frequency} {Domain} {Causal} {Effects} for {Linear} {Processes}},\n\turl = {http://arxiv.org/abs/2406.18191},\n\tdoi = {10.48550/arXiv.2406.18191},\n\tabstract = {Structural vector autoregressive (SVAR) processes are commonly used time series models to identify and quantify causal interactions between dynamically interacting processes from observational data. The causal relationships between these processes can be effectively represented by a finite directed process graph - a graph that connects two processes whenever there is a direct delayed or simultaneous effect between them. Recent research has introduced a framework for quantifying frequency domain causal effects along paths on the process graph. This framework allows to identify how the spectral density of one process is contributing to the spectral density of another. In the current work, we characterise the asymptotic distribution of causal effect and spectral contribution estimators in terms of algebraic relations dictated by the process graph. Based on the asymptotic distribution we construct approximate confidence intervals and Wald type hypothesis tests for the estimated effects and spectral contributions. Under the assumption of causal sufficiency, we consider the class of differentiable estimators for frequency domain causal quantities, and within this class we identify the asymptotically optimal estimator. We illustrate the frequency domain Wald tests and uncertainty approximation on synthetic data, and apply them to analyse the impact of the 10 to 11 year solar cycle on the North Atlantic Oscillation (NAO). Our results confirm a significant effect of the solar cycle on the NAO at the 10 to 11 year time scale.},\n\turldate = {2026-02-26},\n\tpublisher = {arXiv},\n\tauthor = {Reiter, Nicolas-Domenic and Wahl, Jonas and Hegerl, Gabriele C. and Runge, Jakob},\n\tmonth = jun,\n\tyear = {2024},\n\tnote = {arXiv:2406.18191 [stat]},\n\tkeywords = {Mathematics - Statistics Theory, Statistics - Methodology},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Identifying linearly-mixed causal representations from multi-node interventions.\n \n \n \n \n\n\n \n Bing, S.; Ninad, U.; Wahl, J.; and Runge, J.\n\n\n \n\n\n\n In Causal Learning and Reasoning, pages 843–867, 2024. PMLR\n \n\n\n\n
\n\n\n\n \n \n $\"Identifying$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{bingIdentifyingLinearlymixedCausal2024,\n\ttitle = {Identifying linearly-mixed causal representations from multi-node interventions},\n\turl = {https://proceedings.mlr.press/v236/bing24a},\n\turldate = {2026-02-26},\n\tbooktitle = {Causal {Learning} and {Reasoning}},\n\tpublisher = {PMLR},\n\tauthor = {Bing, Simon and Ninad, Urmi and Wahl, Jonas and Runge, Jakob},\n\tyear = {2024},\n\tpages = {843--867},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Causal inference on process graphs, part II: Causal structure and effect identification.\n \n \n \n \n\n\n \n Reiter, N.; Wahl, J.; Gerhardus, A.; and Runge, J.\n\n\n \n\n\n\n arXiv preprint arXiv:2406.17422. 2024.\n \n\n\n\n
\n\n\n\n \n \n $\"Causal$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{reiterCausalInferenceProcess2024,\n\ttitle = {Causal inference on process graphs, part {II}: {Causal} structure and effect identification},\n\tshorttitle = {Causal inference on process graphs, part {II}},\n\turl = {https://projecteuclid.org/journals/bernoulli/volume-32/issue-2/Causal-inference-on-process-graphs--Causal-structure-and-effect/10.3150/25-BEJ1909.short},\n\turldate = {2026-02-26},\n\tjournal = {arXiv preprint arXiv:2406.17422},\n\tauthor = {Reiter, Nicolas-Domenic and Wahl, Jonas and Gerhardus, Andreas and Runge, Jakob},\n\tyear = {2024},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Common European Language Data Space.\n \n \n \n \n\n\n \n Rehm, G.; Piperidis, S.; Choukri, K.; Vasiļjevs, A.; Marheinecke, K.; Arranz, V.; Bērziņš, A.; Deligiannis, M.; Galanis, D.; Giagkou, M.; Gkirtzou, K.; Gkoumas, D.; Grützner-Zahn, A.; Kolovou, A.; Labropoulou, P.; Lagzdiņš, A.; Leitner, E.; Mapelli, V.; Mazo, H.; Ostermann, S.; Racioppa, S.; Rigault, M.; and Voukoutis, L.\n\n\n \n\n\n\n In Calzolari, N.; Kan, M.; Hoste, V.; Lenci, A.; Sakti, S.; and Xue, N., editor(s), Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pages 3579–3586, Torino, Italia, May 2024. ELRA and ICCL\n \n\n\n\n
\n\n\n\n \n \n $\"Common$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{rehmCommonEuropeanLanguage2024,\n\taddress = {Torino, Italia},\n\ttitle = {Common {European} {Language} {Data} {Space}},\n\turl = {https://aclanthology.org/2024.lrec-main.317/},\n\tabstract = {The Common European Language Data Space (LDS) is an integral part of the EU data strategy, which aims at developing a single market for data. Its decentralised technical infrastructure and governance scheme are currently being developed by the LDS project, which also has dedicated tasks for proof-of-concept prototypes, handling legal aspects, raising awareness and promoting the LDS through events and social media channels. The LDS is part of a broader vision for establishing all necessary components to develop European large language models.},\n\turldate = {2025-05-22},\n\tbooktitle = {Proceedings of the 2024 {Joint} {International} {Conference} on {Computational} {Linguistics}, {Language} {Resources} and {Evaluation} ({LREC}-{COLING} 2024)},\n\tpublisher = {ELRA and ICCL},\n\tauthor = {Rehm, Georg and Piperidis, Stelios and Choukri, Khalid and Vasiļjevs, Andrejs and Marheinecke, Katrin and Arranz, Victoria and Bērziņš, Aivars and Deligiannis, Miltos and Galanis, Dimitris and Giagkou, Maria and Gkirtzou, Katerina and Gkoumas, Dimitris and Grützner-Zahn, Annika and Kolovou, Athanasia and Labropoulou, Penny and Lagzdiņš, Andis and Leitner, Elena and Mapelli, Valérie and Mazo, Hélène and Ostermann, Simon and Racioppa, Stefania and Rigault, Mickaël and Voukoutis, Leon},\n\teditor = {Calzolari, Nicoletta and Kan, Min-Yen and Hoste, Veronique and Lenci, Alessandro and Sakti, Sakriani and Xue, Nianwen},\n\tmonth = may,\n\tyear = {2024},\n\tpages = {3579--3586},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Effective Human Oversight of AI-Based Systems: A Signal Detection Perspective on the Detection of Inaccurate and Unfair Outputs.\n \n \n \n\n\n \n Langer, M.; Baum, K.; and Schlicker, N.\n\n\n \n\n\n\n Minds and Machines, 35(1): 1–30. 2024.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n \n \n 1 download\n \n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{langerEffectiveHumanOversight2024,\n\ttitle = {Effective {Human} {Oversight} of {AI}-{Based} {Systems}: {A} {Signal} {Detection} {Perspective} on the {Detection} of {Inaccurate} and {Unfair} {Outputs}},\n\tvolume = {35},\n\tdoi = {10.1007/s11023-024-09701-0},\n\tnumber = {1},\n\tjournal = {Minds and Machines},\n\tauthor = {Langer, Markus and Baum, Kevin and Schlicker, Nadine},\n\tyear = {2024},\n\tpages = {1--30},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Foundations of causal discovery on groups of variables.\n \n \n \n\n\n \n Wahl, J.; Ninad, U.; and Runge, J.\n\n\n \n\n\n\n Journal of Causal Inference, 12(1): 20230041. 2024.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{wahlFoundationsCausalDiscovery2024,\n\ttitle = {Foundations of causal discovery on groups of variables},\n\tvolume = {12},\n\tnumber = {1},\n\tjournal = {Journal of Causal Inference},\n\tauthor = {Wahl, Jonas and Ninad, Urmi and Runge, Jakob},\n\tyear = {2024},\n\tpages = {20230041},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Adapting Multilingual LLMs to Low-Resource Languages with Knowledge Graphs via Adapters.\n \n \n \n \n\n\n \n Gurgurov, D.; Hartmann, M.; and Ostermann, S.\n\n\n \n\n\n\n In Biswas, R.; Kaffee, L.; Agarwal, O.; Minervini, P.; Singh, S.; and de Melo, G., editor(s), Proceedings of the 1st Workshop on Knowledge Graphs and Large Language Models (KaLLM 2024), pages 63–74, Bangkok, Thailand, August 2024. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"Adapting$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{gurgurovAdaptingMultilingualLLMs2024,\n\taddress = {Bangkok, Thailand},\n\ttitle = {Adapting {Multilingual} {LLMs} to {Low}-{Resource} {Languages} with {Knowledge} {Graphs} via {Adapters}},\n\turl = {https://aclanthology.org/2024.kallm-1.7/},\n\tdoi = {10.18653/v1/2024.kallm-1.7},\n\tabstract = {This paper explores the integration of graph knowledge from linguistic ontologies into multilingual Large Language Models (LLMs) using adapters to improve performance for low-resource languages (LRLs) in sentiment analysis (SA) and named entity recognition (NER). Building upon successful parameter-efficient fine-tuning techniques, such as K-ADAPTER and MAD-X, we propose a similar approach for incorporating knowledge from multilingual graphs, connecting concepts in various languages with each other through linguistic relationships, into multilingual LLMs for LRLs. Specifically, we focus on eight LRLs — Maltese, Bulgarian, Indonesian, Nepali, Javanese, Uyghur, Tibetan, and Sinhala — and employ language-specific adapters fine-tuned on data extracted from the language-specific section of ConceptNet, aiming to enable knowledge transfer across the languages covered by the knowledge graph. We compare various fine-tuning objectives, including standard Masked Language Modeling (MLM), MLM with full-word masking, and MLM with targeted masking, to analyze their effectiveness in learning and integrating the extracted graph data. Through empirical evaluation on language-specific tasks, we assess how structured graph knowledge affects the performance of multilingual LLMs for LRLs in SA and NER, providing insights into the potential benefits of adapting language models for low-resource scenarios.},\n\turldate = {2025-05-22},\n\tbooktitle = {Proceedings of the 1st {Workshop} on {Knowledge} {Graphs} and {Large} {Language} {Models} ({KaLLM} 2024)},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Gurgurov, Daniil and Hartmann, Mareike and Ostermann, Simon},\n\teditor = {Biswas, Russa and Kaffee, Lucie-Aimée and Agarwal, Oshin and Minervini, Pasquale and Singh, Sameer and de Melo, Gerard},\n\tmonth = aug,\n\tyear = {2024},\n\tpages = {63--74},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n HybridBERT - Making BERT Pretraining More Efficient Through Hybrid Mixture of Attention Mechanisms.\n \n \n \n \n\n\n \n Srinivasagan, G.; and Ostermann, S.\n\n\n \n\n\n\n In Cao, Y. (.; Papadimitriou, I.; Ovalle, A.; Zampieri, M.; Ferraro, F.; and Swayamdipta, S., editor(s), Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 4: Student Research Workshop), pages 285–291, Mexico City, Mexico, June 2024. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"HybridBERT$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{srinivasaganHybridBERTMakingBERT2024,\n\taddress = {Mexico City, Mexico},\n\ttitle = {{HybridBERT} - {Making} {BERT} {Pretraining} {More} {Efficient} {Through} {Hybrid} {Mixture} of {Attention} {Mechanisms}},\n\turl = {https://aclanthology.org/2024.naacl-srw.30/},\n\tdoi = {10.18653/v1/2024.naacl-srw.30},\n\tabstract = {Pretrained transformer-based language models have produced state-of-the-art performance in most natural language understanding tasks. These models undergo two stages of training: pretraining on a huge corpus of data and fine-tuning on a specific downstream task. The pretraining phase is extremely compute-intensive and requires several high-performance computing devices like GPUs and several days or even months of training, but it is crucial for the model to capture global knowledge and also has a significant impact on the fine-tuning task. This is a major roadblock for researchers without access to sophisticated computing resources. To overcome this challenge, we propose two novel hybrid architectures called HybridBERT (HBERT), which combine self-attention and additive attention mechanisms together with sub-layer normalization. We introduce a computing budget to the pretraining phase, limiting the training time and usage to a single GPU. We show that HBERT attains twice the pretraining accuracy of a vanilla-BERT baseline. We also evaluate our proposed models on two downstream tasks, where we outperform BERT-base while accelerating inference. Moreover, we study the effect of weight initialization with a limited pretraining budget. The code and models are publicly available at: www.github.com/gokulsg/HBERT/.},\n\turldate = {2025-05-22},\n\tbooktitle = {Proceedings of the 2024 {Conference} of the {North} {American} {Chapter} of the {Association} for {Computational} {Linguistics}: {Human} {Language} {Technologies} ({Volume} 4: {Student} {Research} {Workshop})},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Srinivasagan, Gokul and Ostermann, Simon},\n\teditor = {Cao, Yang (Trista) and Papadimitriou, Isabel and Ovalle, Anaelia and Zampieri, Marcos and Ferraro, Francis and Swayamdipta, Swabha},\n\tmonth = jun,\n\tyear = {2024},\n\tpages = {285--291},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n CoXQL: A Dataset for Parsing Explanation Requests in Conversational XAI Systems.\n \n \n \n \n\n\n \n Wang, Q.; Anikina, T.; Feldhus, N.; Ostermann, S.; and Möller, S.\n\n\n \n\n\n\n In Al-Onaizan, Y.; Bansal, M.; and Chen, Y., editor(s), Findings of the Association for Computational Linguistics: EMNLP 2024, pages 1410–1422, Miami, Florida, USA, November 2024. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"CoXQL:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{wangCoXQLDatasetParsing2024,\n\taddress = {Miami, Florida, USA},\n\ttitle = {{CoXQL}: {A} {Dataset} for {Parsing} {Explanation} {Requests} in {Conversational} {XAI} {Systems}},\n\tshorttitle = {{CoXQL}},\n\turl = {https://aclanthology.org/2024.findings-emnlp.76/},\n\tdoi = {10.18653/v1/2024.findings-emnlp.76},\n\tabstract = {Conversational explainable artificial intelligence (ConvXAI) systems based on large language models (LLMs) have garnered significant interest from the research community in natural language processing (NLP) and human-computer interaction (HCI). Such systems can provide answers to user questions about explanations in dialogues, have the potential to enhance users' comprehension and offer more information about the decision-making and generation processes of LLMs. Currently available ConvXAI systems are based on intent recognition rather than free chat, as this has been found to be more precise and reliable in identifying users' intentions. However, the recognition of intents still presents a challenge in the case of ConvXAI, since little training data exist and the domain is highly specific, as there is a broad range of XAI methods to map requests onto. In order to bridge this gap, we present CoXQL, the first dataset in the NLP domain for user intent recognition in ConvXAI, covering 31 intents, seven of which require filling multiple slots. Subsequently, we enhance an existing parsing approach by incorporating template validations, and conduct an evaluation of several LLMs on CoXQL using different parsing strategies. We conclude that the improved parsing approach (MP+) surpasses the performance of previous approaches. We also discover that intents with multiple slots remain highly challenging for LLMs.},\n\turldate = {2025-05-22},\n\tbooktitle = {Findings of the {Association} for {Computational} {Linguistics}: {EMNLP} 2024},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Wang, Qianli and Anikina, Tatiana and Feldhus, Nils and Ostermann, Simon and Möller, Sebastian},\n\teditor = {Al-Onaizan, Yaser and Bansal, Mohit and Chen, Yun-Nung},\n\tmonth = nov,\n\tyear = {2024},\n\tpages = {1410--1422},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n AI Engineering for Trust by Design.\n \n \n \n \n\n\n \n Meyer-Vitali, A.\n\n\n \n\n\n\n In Proceedings of the 12th International Conference on Model-Based Software and Systems Engineering (MODELSWARD 2024), pages 357–364, Rome, Italy, February 2024. SCITEPRESS – Science and Technology Publications, Lda\n \n\n\n\n
\n\n\n\n \n \n $\"AI$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{meyer-vitaliAIEngineeringTrust2024a,\n\taddress = {Rome, Italy},\n\ttitle = {{AI} {Engineering} for {Trust} by {Design}},\n\tcopyright = {All rights reserved},\n\tisbn = {978-989-758-682-8},\n\turl = {https://www.scitepress.org/PublicationsDetail.aspx?ID=/skO/EwOJr4=&t=1},\n\tdoi = {10.5220/0012622400003645},\n\tabstract = {Digital Library},\n\turldate = {2024-02-29},\n\tbooktitle = {Proceedings of the 12th {International} {Conference} on {Model}-{Based} {Software} and {Systems} {Engineering} ({MODELSWARD} 2024)},\n\tpublisher = {SCITEPRESS – Science and Technology Publications, Lda},\n\tauthor = {Meyer-Vitali, André},\n\tmonth = feb,\n\tyear = {2024},\n\tpages = {357--364},\n}\n\n\n\n\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n MMAR: Multilingual and Multimodal Anaphora Resolution in Instructional Videos.\n \n \n \n \n\n\n \n Oguz, C.; Denis, P.; Ostermann, S.; Vincent, E.; Skachkova, N.; and Genabith, J. V.\n\n\n \n\n\n\n In Al-Onaizan, Y.; Bansal, M.; and Chen, Y., editor(s), Findings of the Association for Computational Linguistics: EMNLP 2024, pages 1618–1633, Miami, Florida, USA, November 2024. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"MMAR:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{oguzMMARMultilingualMultimodal2024a,\n\taddress = {Miami, Florida, USA},\n\ttitle = {{MMAR}: {Multilingual} and {Multimodal} {Anaphora} {Resolution} in {Instructional} {Videos}},\n\tshorttitle = {{MMAR}},\n\turl = {https://aclanthology.org/2024.findings-emnlp.88/},\n\tdoi = {10.18653/v1/2024.findings-emnlp.88},\n\tabstract = {Multilingual anaphora resolution identifies referring expressions and implicit arguments in texts and links to antecedents that cover several languages. In the most challenging setting, cross-lingual anaphora resolution, training data, and test data are in different languages. As knowledge needs to be transferred across languages, this task is challenging, both in the multilingual and cross-lingual setting. We hypothesize that one way to alleviate some of the difficulty of the task is to include multimodal information in the form of images (i.e. frames extracted from instructional videos). Such visual inputs are by nature language agnostic, therefore cross- and multilingual anaphora resolution should benefit from visual information. In this paper, we provide the first multilingual and multimodal dataset annotated with anaphoric relations and present experimental results for end-to-end multimodal and multilingual anaphora resolution. Given gold mentions, multimodal features improve anaphora resolution results by {\\textbackslash}textbackslashtextasciitilde10 \\% for unseen languages.},\n\turldate = {2025-05-22},\n\tbooktitle = {Findings of the {Association} for {Computational} {Linguistics}: {EMNLP} 2024},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Oguz, Cennet and Denis, Pascal and Ostermann, Simon and Vincent, Emmanuel and Skachkova, Natalia and Genabith, Josef Van},\n\teditor = {Al-Onaizan, Yaser and Bansal, Mohit and Chen, Yun-Nung},\n\tmonth = nov,\n\tyear = {2024},\n\tpages = {1618--1633},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n A Comparison of Different Tokenization Methods for the Georgian Language.\n \n \n \n \n\n\n \n Mikaberidze, B.; Saghinadze, T.; Mikaberidze, G.; Kalandadze, R.; Pkhakadze, K.; van Genabith, J.; Ostermann, S.; van der Plas, L.; and Müller, P.\n\n\n \n\n\n\n In Abbas, M.; and Freihat, A. A., editor(s), Proceedings of the 7th International Conference on Natural Language and Speech Processing (ICNLSP 2024), pages 199–208, Trento, October 2024. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"A$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{mikaberidzeComparisonDifferentTokenization2024,\n\taddress = {Trento},\n\ttitle = {A {Comparison} of {Different} {Tokenization} {Methods} for the {Georgian} {Language}},\n\turl = {https://aclanthology.org/2024.icnlsp-1.22/},\n\turldate = {2025-05-22},\n\tbooktitle = {Proceedings of the 7th {International} {Conference} on {Natural} {Language} and {Speech} {Processing} ({ICNLSP} 2024)},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Mikaberidze, Beso and Saghinadze, Temo and Mikaberidze, Guram and Kalandadze, Raphael and Pkhakadze, Konstantine and van Genabith, Josef and Ostermann, Simon and van der Plas, Lonneke and Müller, Philipp},\n\teditor = {Abbas, Mourad and Freihat, Abed Alhakim},\n\tmonth = oct,\n\tyear = {2024},\n\tpages = {199--208},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Invariance & Causal Representation Learning: Prospects and Limitations.\n \n \n \n\n\n \n Bing, S.; Hochsprung, T.; Wahl, J.; Ninad, U.; and Runge, J.\n\n\n \n\n\n\n Transactions of Machine Learning Research (TMLR). ISSN: 2835-8856., (https://openreview.net/forum?id=lpOC6s4B). 2024.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{bingInvarianceCausalRepresentation2024,\n\ttitle = {Invariance \\& {Causal} {Representation} {Learning}: {Prospects} and {Limitations}},\n\tnumber = {https://openreview.net/forum?id=lpOC6s4B},\n\tjournal = {Transactions of Machine Learning Research (TMLR). ISSN: 2835-8856.},\n\tauthor = {Bing, Simon and Hochsprung, Tom and Wahl, Jonas and Ninad, Urmi and Runge, Jakob},\n\tyear = {2024},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Causal Inference for Spatial Data Analytics (Dagstuhl Seminar 24202).\n \n \n \n \n\n\n \n Tomko, M.; Xin, Y.; and Wahl, J.\n\n\n \n\n\n\n Dagstuhl Reports, 14(5): 25–57. 2024.\n \n\n\n\n
\n\n\n\n \n \n $\"Causal$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{tomkoCausalInferenceSpatial2024,\n\ttitle = {Causal {Inference} for {Spatial} {Data} {Analytics} ({Dagstuhl} {Seminar} 24202)},\n\tvolume = {14},\n\tissn = {2192-5283},\n\turl = {https://drops.dagstuhl.de/entities/document/10.4230/DagRep.14.5.25},\n\tdoi = {10.4230/DagRep.14.5.25},\n\tnumber = {5},\n\tjournal = {Dagstuhl Reports},\n\tauthor = {Tomko, Martin and Xin, Yanan and Wahl, Jonas},\n\teditor = {Tomko, Martin and Xin, Yanan and Wahl, Jonas},\n\tyear = {2024},\n\tpages = {25--57},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Divergent Token Metrics: Measuring degradation to prune away LLM components – and optimize quantization.\n \n \n \n \n\n\n \n Deiseroth, B.; Meuer, M.; Gritsch, N.; Eichenberg, C.; Schramowski, P.; Aßenmacher, M.; and Kersting, K.\n\n\n \n\n\n\n In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics, 2024. arXiv\n \n\n\n\n
\n\n\n\n \n \n $\"Divergent$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n\n\n\n

@inproceedings{deiserothDivergentTokenMetrics2024,\n\ttitle = {Divergent {Token} {Metrics}: {Measuring} degradation to prune away {LLM} components -- and optimize quantization},\n\tcopyright = {Creative Commons Attribution 4.0 International},\n\tshorttitle = {Divergent {Token} {Metrics}},\n\turl = {https://arxiv.org/abs/2311.01544},\n\tdoi = {10.48550/ARXIV.2311.01544},\n\tabstract = {Large Language Models (LLMs) have reshaped natural language processing with their impressive capabilities. However, their ever-increasing size has raised concerns about their effective deployment and the need for LLM compression. This study introduces the Divergent Token Metrics (DTMs), a novel approach to assessing compressed LLMs, addressing the limitations of traditional perplexity or accuracy measures that fail to accurately reflect text generation quality. DTMs measure token divergences that allow deeper insights into the subtleties of model compression, in particular, when evaluating components' impacts individually. Utilizing the First Divergent Token Metric (FDTM) in model sparsification reveals that 25\\% of all attention components can be pruned beyond 90\\% on the Llama-2 model family, still keeping SOTA performance. For quantization, FDTM suggests that more than 80\\% of parameters can be naively transformed to int8 without special outlier management. These evaluations indicate the necessity of choosing appropriate compressions for parameters individually -- and that FDTM can identify those -- while standard metrics result in deteriorated outcomes.},\n\turldate = {2025-05-21},\n\tbooktitle = {Proceedings of the 2024 {Conference} of the {North} {American} {Chapter} of the {Association} for {Computational} {Linguistics}},\n\tpublisher = {arXiv},\n\tauthor = {Deiseroth, Björn and Meuer, Max and Gritsch, Nikolas and Eichenberg, Constantin and Schramowski, Patrick and Aßenmacher, Matthias and Kersting, Kristian},\n\tyear = {2024},\n\tkeywords = {Computation and Language (cs.CL), FOS: Computer and information sciences, Machine Learning (cs.LG)},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Acting for the Right Reasons: Creating Reason-Sensitive Artificial Moral Agents.\n \n \n \n \n\n\n \n Baum, K.; Dargasz, L.; Jahn, F.; Gros, T. P.; and Wolf, V.\n\n\n \n\n\n\n In 2024. \n \n\n\n\n
\n\n\n\n \n \n $\"Acting$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n \n \n 1 download\n \n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{baumActingRightReasons2024,\n\ttitle = {Acting for the {Right} {Reasons}: {Creating} {Reason}-{Sensitive} {Artificial} {Moral} {Agents}},\n\turl = {https://openreview.net/forum?id=Mgyr7EY7X9&referrer=%5Bthe%20profile%20of%20Kevin%20Baum%5D(%2Fprofile%3Fid%3D~Kevin_Baum1)},\n\tdoi = {https://doi.org/10.48550/arXiv.2409.15014},\n\tauthor = {Baum, Kevin and Dargasz, Lisa and Jahn, Felix and Gros, Timo P. and Wolf, Verena},\n\tyear = {2024},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n On the Quest for Effectiveness in Human Oversight: Interdisciplinary Perspectives.\n \n \n \n \n\n\n \n Sterz, S.; Baum, K.; Biewer, S.; Hermanns, H.; Lauber-Rönsberg, A.; Meinel, P.; and Langer, M.\n\n\n \n\n\n\n In The 2024 ACM Conference on Fairness, Accountability, and Transparency, pages 2495–2507, Rio de Janeiro Brazil, June 2024. ACM\n \n\n\n\n
\n\n\n\n \n \n $\"On$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{sterzQuestEffectivenessHuman2024,\n\taddress = {Rio de Janeiro Brazil},\n\ttitle = {On the {Quest} for {Effectiveness} in {Human} {Oversight}: {Interdisciplinary} {Perspectives}},\n\tisbn = {979-8-4007-0450-5},\n\tshorttitle = {On the {Quest} for {Effectiveness} in {Human} {Oversight}},\n\turl = {https://dl.acm.org/doi/10.1145/3630106.3659051},\n\tdoi = {10.1145/3630106.3659051},\n\tlanguage = {en},\n\turldate = {2025-05-13},\n\tbooktitle = {The 2024 {ACM} {Conference} on {Fairness}, {Accountability}, and {Transparency}},\n\tpublisher = {ACM},\n\tauthor = {Sterz, Sarah and Baum, Kevin and Biewer, Sebastian and Hermanns, Holger and Lauber-Rönsberg, Anne and Meinel, Philip and Langer, Markus},\n\tmonth = jun,\n\tyear = {2024},\n\tpages = {2495--2507},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Software doping analysis for human oversight.\n \n \n \n \n\n\n \n Biewer, S.; Baum, K.; Sterz, S.; Hermanns, H.; Hetmank, S.; Langer, M.; Lauber-Rönsberg, A.; and Lehr, F.\n\n\n \n\n\n\n Formal Methods in System Design. April 2024.\n \n\n\n\n
\n\n\n\n \n \n $\"Software$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n \n \n 4 downloads\n \n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{biewerSoftwareDopingAnalysis2024,\n\ttitle = {Software doping analysis for human oversight},\n\tissn = {0925-9856, 1572-8102},\n\turl = {https://link.springer.com/10.1007/s10703-024-00445-2},\n\tdoi = {10.1007/s10703-024-00445-2},\n\tabstract = {Abstract \n             \n              This article introduces a framework that is meant to assist in mitigating societal risks that software can pose. Concretely, this encompasses facets of software doping as well as unfairness and discrimination in high-risk decision-making systems. The term \n              software doping \n              refers to software that contains surreptitiously added functionality that is against the interest of the user. A prominent example of software doping are the tampered emission cleaning systems that were found in millions of cars around the world when the diesel emissions scandal surfaced. The first part of this article combines the formal foundations of software doping analysis with established probabilistic falsification techniques to arrive at a black-box analysis technique for identifying undesired effects of software. We apply this technique to emission cleaning systems in diesel cars but also to high-risk systems that evaluate humans in a possibly unfair or discriminating way. We demonstrate how our approach can assist humans-in-the-loop to make better informed and more responsible decisions. This is to promote effective human oversight, which will be a central requirement enforced by the European Union’s upcoming AI Act. We complement our technical contribution with a juridically, philosophically, and psychologically informed perspective on the potential problems caused by such systems.},\n\tlanguage = {en},\n\turldate = {2025-05-13},\n\tjournal = {Formal Methods in System Design},\n\tauthor = {Biewer, Sebastian and Baum, Kevin and Sterz, Sarah and Hermanns, Holger and Hetmank, Sven and Langer, Markus and Lauber-Rönsberg, Anne and Lehr, Franz},\n\tmonth = apr,\n\tyear = {2024},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Task Prompt Vectors: Effective Initialization through Multi-Task Soft-Prompt Transfer.\n \n \n \n \n\n\n \n Belanec, R.; Ostermann, S.; Srba, I.; and Bielikova, M.\n\n\n \n\n\n\n 2024.\n \n\n\n\n
\n\n\n\n \n \n $\"Task$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n\n\n\n

@misc{belanecTaskPromptVectors2024,\n\ttitle = {Task {Prompt} {Vectors}: {Effective} {Initialization} through {Multi}-{Task} {Soft}-{Prompt} {Transfer}},\n\tcopyright = {Creative Commons Attribution 4.0 International},\n\tshorttitle = {Task {Prompt} {Vectors}},\n\turl = {https://arxiv.org/abs/2408.01119},\n\tdoi = {10.48550/ARXIV.2408.01119},\n\tabstract = {Prompt tuning is an efficient solution for training large language models (LLMs). However, current soft-prompt-based methods often sacrifice multi-task modularity, requiring the training process to be fully or partially repeated for each newly added task. While recent work on task vectors applied arithmetic operations on full model weights to achieve the desired multi-task performance, a similar approach for soft-prompts is still missing. To this end, we introduce Task Prompt Vectors, created by element-wise difference between weights of tuned soft-prompts and their random initialization. Experimental results on 12 NLU datasets show that task prompt vectors can be used in low-resource settings to effectively initialize prompt tuning on similar tasks. In addition, we show that task prompt vectors are independent of the random initialization of prompt tuning on 2 different language model architectures. This allows prompt arithmetics with the pre-trained vectors from different tasks. In this way, we provide a competitive alternative to state-of-the-art baselines by arithmetic addition of task prompt vectors from multiple tasks.},\n\turldate = {2025-05-13},\n\tpublisher = {arXiv},\n\tauthor = {Belanec, Robert and Ostermann, Simon and Srba, Ivan and Bielikova, Maria},\n\tyear = {2024},\n\tkeywords = {Computation and Language (cs.CL), FOS: Computer and information sciences},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Motion Primitives as the Action Space of Deep Q-Learning for Planning in Autonomous Driving.\n \n \n \n \n\n\n \n Schneider, T.; Pedrosa, M. V. A.; Gros, T. P.; Wolf, V.; and Flaßkamp, K.\n\n\n \n\n\n\n IEEE Transactions on Intelligent Transportation Systems, 25(11): 17852–17864. November 2024.\n \n\n\n\n
\n\n\n\n \n \n $\"Motion$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{schneiderMotionPrimitivesAction2024,\n\ttitle = {Motion {Primitives} as the {Action} {Space} of {Deep} {Q}-{Learning} for {Planning} in {Autonomous} {Driving}},\n\tvolume = {25},\n\tcopyright = {https://ieeexplore.ieee.org/Xplorehelp/downloads/license-information/IEEE.html},\n\tissn = {1524-9050, 1558-0016},\n\turl = {https://ieeexplore.ieee.org/document/10693315/},\n\tdoi = {10.1109/TITS.2024.3436530},\n\tnumber = {11},\n\turldate = {2025-05-13},\n\tjournal = {IEEE Transactions on Intelligent Transportation Systems},\n\tauthor = {Schneider, Tristan and Pedrosa, Matheus V. A. and Gros, Timo P. and Wolf, Verena and Flaßkamp, Kathrin},\n\tmonth = nov,\n\tyear = {2024},\n\tpages = {17852--17864},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Comparing State-of-the-art Graph Neural Networks and Transformers for General Policy Learning.\n \n \n \n\n\n \n Müller, N. J.; Sánchez, P.; Hoffmann, J.; Wolf, V.; and Gros, T. P.\n\n\n \n\n\n\n . 2024.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{mullerComparingStateoftheartGraph2024,\n\ttitle = {Comparing {State}-of-the-art {Graph} {Neural} {Networks} and {Transformers} for {General} {Policy} {Learning}},\n\tauthor = {Müller, Nicola J. and Sánchez, Pablo and Hoffmann, Jörg and Wolf, Verena and Gros, Timo P.},\n\tyear = {2024},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Occiglot at WMT24: European Open-source Large Language Models Evaluated on Translation.\n \n \n \n \n\n\n \n Avramidis, E.; Grützner-Zahn, A.; Brack, M.; Schramowski, P.; Ortiz Suarez, P.; Ostendorff, M.; Barth, F.; Manakhimova, S.; Macketanz, V.; Rehm, G.; and Kersting, K.\n\n\n \n\n\n\n In Proceedings of the Ninth Conference on Machine Translation, pages 292–298, Miami, Florida, USA, 2024. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"Occiglot$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{avramidisOcciglotWMT24European2024,\n\taddress = {Miami, Florida, USA},\n\ttitle = {Occiglot at {WMT24}: {European} {Open}-source {Large} {Language} {Models} {Evaluated} on {Translation}},\n\tshorttitle = {Occiglot at {WMT24}},\n\turl = {https://aclanthology.org/2024.wmt-1.23},\n\tdoi = {10.18653/v1/2024.wmt-1.23},\n\tlanguage = {en},\n\turldate = {2025-05-13},\n\tbooktitle = {Proceedings of the {Ninth} {Conference} on {Machine} {Translation}},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Avramidis, Eleftherios and Grützner-Zahn, Annika and Brack, Manuel and Schramowski, Patrick and Ortiz Suarez, Pedro and Ostendorff, Malte and Barth, Fabio and Manakhimova, Shushen and Macketanz, Vivien and Rehm, Georg and Kersting, Kristian},\n\tyear = {2024},\n\tpages = {292--298},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n T-FREE: Subword Tokenizer-Free Generative LLMs via Sparse Representations for Memory-Efficient Embeddings.\n \n \n \n \n\n\n \n Deiseroth, B.; Brack, M.; Schramowski, P.; Kersting, K.; and Weinbach, S.\n\n\n \n\n\n\n In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 21829–21851, Miami, Florida, USA, 2024. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"T-FREE:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{deiserothTFREESubwordTokenizerFree2024,\n\taddress = {Miami, Florida, USA},\n\ttitle = {T-{FREE}: {Subword} {Tokenizer}-{Free} {Generative} {LLMs} via {Sparse} {Representations} for {Memory}-{Efficient} {Embeddings}},\n\tshorttitle = {T-{FREE}},\n\turl = {https://aclanthology.org/2024.emnlp-main.1217},\n\tdoi = {10.18653/v1/2024.emnlp-main.1217},\n\tlanguage = {en},\n\turldate = {2025-05-13},\n\tbooktitle = {Proceedings of the 2024 {Conference} on {Empirical} {Methods} in {Natural} {Language} {Processing}},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Deiseroth, Björn and Brack, Manuel and Schramowski, Patrick and Kersting, Kristian and Weinbach, Samuel},\n\tyear = {2024},\n\tpages = {21829--21851},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Introducing v0.5 of the AI Safety Benchmark from MLCommons.\n \n \n \n \n\n\n \n Vidgen, B.; Agrawal, A.; Ahmed, A. M.; Akinwande, V.; Al-Nuaimi, N.; Alfaraj, N.; Alhajjar, E.; Aroyo, L.; Bavalatti, T.; Bartolo, M.; Blili-Hamelin, B.; Bollacker, K.; Bomassani, R.; Boston, M. F.; Campos, S.; Chakra, K.; Chen, C.; Coleman, C.; Coudert, Z. D.; Derczynski, L.; Dutta, D.; Eisenberg, I.; Ezick, J.; Frase, H.; Fuller, B.; Gandikota, R.; Gangavarapu, A.; Gangavarapu, A.; Gealy, J.; Ghosh, R.; Goel, J.; Gohar, U.; Goswami, S.; Hale, S. A.; Hutiri, W.; Imperial, J. M.; Jandial, S.; Judd, N.; Juefei-Xu, F.; Khomh, F.; Kailkhura, B.; Kirk, H. R.; Klyman, K.; Knotz, C.; Kuchnik, M.; Kumar, S. H.; Kumar, S.; Lengerich, C.; Li, B.; Liao, Z.; Long, E. P.; Lu, V.; Luger, S.; Mai, Y.; Mammen, P. M.; Manyeki, K.; McGregor, S.; Mehta, V.; Mohammed, S.; Moss, E.; Nachman, L.; Naganna, D. J.; Nikanjam, A.; Nushi, B.; Oala, L.; Orr, I.; Parrish, A.; Patlak, C.; Pietri, W.; Poursabzi-Sangdeh, F.; Presani, E.; Puletti, F.; Röttger, P.; Sahay, S.; Santos, T.; Scherrer, N.; Sebag, A. S.; Schramowski, P.; Shahbazi, A.; Sharma, V.; Shen, X.; Sistla, V.; Tang, L.; Testuggine, D.; Thangarasa, V.; Watkins, E. A.; Weiss, R.; Welty, C.; Wilbers, T.; Williams, A.; Wu, C.; Yadav, P.; Yang, X.; Zeng, Y.; Zhang, W.; Zhdanov, F.; Zhu, J.; Liang, P.; Mattson, P.; and Vanschoren, J.\n\n\n \n\n\n\n 2024.\n \n\n\n\n
\n\n\n\n \n \n $\"Introducing$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n\n\n\n

@misc{vidgenIntroducingV05AI2024,\n\ttitle = {Introducing v0.5 of the {AI} {Safety} {Benchmark} from {MLCommons}},\n\tcopyright = {Creative Commons Attribution 4.0 International},\n\turl = {https://arxiv.org/abs/2404.12241},\n\tdoi = {10.48550/ARXIV.2404.12241},\n\tabstract = {This paper introduces v0.5 of the AI Safety Benchmark, which has been created by the MLCommons AI Safety Working Group. The AI Safety Benchmark has been designed to assess the safety risks of AI systems that use chat-tuned language models. We introduce a principled approach to specifying and constructing the benchmark, which for v0.5 covers only a single use case (an adult chatting to a general-purpose assistant in English), and a limited set of personas (i.e., typical users, malicious users, and vulnerable users). We created a new taxonomy of 13 hazard categories, of which 7 have tests in the v0.5 benchmark. We plan to release version 1.0 of the AI Safety Benchmark by the end of 2024. The v1.0 benchmark will provide meaningful insights into the safety of AI systems. However, the v0.5 benchmark should not be used to assess the safety of AI systems. We have sought to fully document the limitations, flaws, and challenges of v0.5. This release of v0.5 of the AI Safety Benchmark includes (1) a principled approach to specifying and constructing the benchmark, which comprises use cases, types of systems under test (SUTs), language and context, personas, tests, and test items; (2) a taxonomy of 13 hazard categories with definitions and subcategories; (3) tests for seven of the hazard categories, each comprising a unique set of test items, i.e., prompts. There are 43,090 test items in total, which we created with templates; (4) a grading system for AI systems against the benchmark; (5) an openly available platform, and downloadable tool, called ModelBench that can be used to evaluate the safety of AI systems on the benchmark; (6) an example evaluation report which benchmarks the performance of over a dozen openly available chat-tuned language models; (7) a test specification for the benchmark.},\n\turldate = {2025-05-13},\n\tpublisher = {arXiv},\n\tauthor = {Vidgen, Bertie and Agrawal, Adarsh and Ahmed, Ahmed M. and Akinwande, Victor and Al-Nuaimi, Namir and Alfaraj, Najla and Alhajjar, Elie and Aroyo, Lora and Bavalatti, Trupti and Bartolo, Max and Blili-Hamelin, Borhane and Bollacker, Kurt and Bomassani, Rishi and Boston, Marisa Ferrara and Campos, Siméon and Chakra, Kal and Chen, Canyu and Coleman, Cody and Coudert, Zacharie Delpierre and Derczynski, Leon and Dutta, Debojyoti and Eisenberg, Ian and Ezick, James and Frase, Heather and Fuller, Brian and Gandikota, Ram and Gangavarapu, Agasthya and Gangavarapu, Ananya and Gealy, James and Ghosh, Rajat and Goel, James and Gohar, Usman and Goswami, Sujata and Hale, Scott A. and Hutiri, Wiebke and Imperial, Joseph Marvin and Jandial, Surgan and Judd, Nick and Juefei-Xu, Felix and Khomh, Foutse and Kailkhura, Bhavya and Kirk, Hannah Rose and Klyman, Kevin and Knotz, Chris and Kuchnik, Michael and Kumar, Shachi H. and Kumar, Srijan and Lengerich, Chris and Li, Bo and Liao, Zeyi and Long, Eileen Peters and Lu, Victor and Luger, Sarah and Mai, Yifan and Mammen, Priyanka Mary and Manyeki, Kelvin and McGregor, Sean and Mehta, Virendra and Mohammed, Shafee and Moss, Emanuel and Nachman, Lama and Naganna, Dinesh Jinenhally and Nikanjam, Amin and Nushi, Besmira and Oala, Luis and Orr, Iftach and Parrish, Alicia and Patlak, Cigdem and Pietri, William and Poursabzi-Sangdeh, Forough and Presani, Eleonora and Puletti, Fabrizio and Röttger, Paul and Sahay, Saurav and Santos, Tim and Scherrer, Nino and Sebag, Alice Schoenauer and Schramowski, Patrick and Shahbazi, Abolfazl and Sharma, Vin and Shen, Xudong and Sistla, Vamsi and Tang, Leonard and Testuggine, Davide and Thangarasa, Vithursan and Watkins, Elizabeth Anne and Weiss, Rebecca and Welty, Chris and Wilbers, Tyler and Williams, Adina and Wu, Carole-Jean and Yadav, Poonam and Yang, Xianjun and Zeng, Yi and Zhang, Wenhui and Zhdanov, Fedor and Zhu, Jiacheng and Liang, Percy and Mattson, Peter and Vanschoren, Joaquin},\n\tyear = {2024},\n\tkeywords = {Artificial Intelligence (cs.AI), Computation and Language (cs.CL), FOS: Computer and information sciences},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Community OSCAR: A Community Effort for Multilingual Web Data.\n \n \n \n \n\n\n \n Brack, M.; Ostendorff, M.; Ortiz Suarez, P.; Saiz, J. J.; Castilla, I. L.; Palomar-Giner, J.; Shvets, A.; Schramowski, P.; Rehm, G.; Villegas, M.; and Kersting, K.\n\n\n \n\n\n\n In Proceedings of the Fourth Workshop on Multilingual Representation Learning (MRL 2024), pages 232–235, Miami, Florida, USA, 2024. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"Community$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{brackCommunityOSCARCommunity2024,\n\taddress = {Miami, Florida, USA},\n\ttitle = {Community {OSCAR}: {A} {Community} {Effort} for {Multilingual} {Web} {Data}},\n\tshorttitle = {Community {OSCAR}},\n\turl = {https://aclanthology.org/2024.mrl-1.19},\n\tdoi = {10.18653/v1/2024.mrl-1.19},\n\tlanguage = {en},\n\turldate = {2025-05-13},\n\tbooktitle = {Proceedings of the {Fourth} {Workshop} on {Multilingual} {Representation} {Learning} ({MRL} 2024)},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Brack, Manuel and Ostendorff, Malte and Ortiz Suarez, Pedro and Saiz, José Javier and Castilla, Iñaki Lacunza and Palomar-Giner, Jorge and Shvets, Alexander and Schramowski, Patrick and Rehm, Georg and Villegas, Marta and Kersting, Kristian},\n\tyear = {2024},\n\tpages = {232--235},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Core Tokensets for Data-efficient Sequential Training of Transformers.\n \n \n \n \n\n\n \n Paul, S.; Brack, M.; Schramowski, P.; Kersting, K.; and Mundt, M.\n\n\n \n\n\n\n 2024.\n \n\n\n\n
\n\n\n\n \n \n $\"Core$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n\n\n\n

@misc{paulCoreTokensetsDataefficient2024,\n\ttitle = {Core {Tokensets} for {Data}-efficient {Sequential} {Training} of {Transformers}},\n\tcopyright = {Creative Commons Attribution Non Commercial Share Alike 4.0 International},\n\turl = {https://arxiv.org/abs/2410.05800},\n\tdoi = {10.48550/ARXIV.2410.05800},\n\tabstract = {Deep networks are frequently tuned to novel tasks and continue learning from ongoing data streams. Such sequential training requires consolidation of new and past information, a challenge predominantly addressed by retaining the most important data points - formally known as coresets. Traditionally, these coresets consist of entire samples, such as images or sentences. However, recent transformer architectures operate on tokens, leading to the famous assertion that an image is worth 16x16 words. Intuitively, not all of these tokens are equally informative or memorable. Going beyond coresets, we thus propose to construct a deeper-level data summary on the level of tokens. Our respectively named core tokensets both select the most informative data points and leverage feature attribution to store only their most relevant features. We demonstrate that core tokensets yield significant performance retention in incremental image classification, open-ended visual question answering, and continual image captioning with significantly reduced memory. In fact, we empirically find that a core tokenset of 1{\\textbackslash}\\% of the data performs comparably to at least a twice as large and up to 10 times larger coreset.},\n\turldate = {2025-05-13},\n\tpublisher = {arXiv},\n\tauthor = {Paul, Subarnaduti and Brack, Manuel and Schramowski, Patrick and Kersting, Kristian and Mundt, Martin},\n\tyear = {2024},\n\tkeywords = {Artificial Intelligence (cs.AI), Computer Vision and Pattern Recognition (cs.CV), FOS: Computer and information sciences},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Auditing and instructing text-to-image generation models on fairness.\n \n \n \n \n\n\n \n Friedrich, F.; Brack, M.; Struppek, L.; Hintersdorf, D.; Schramowski, P.; Luccioni, S.; and Kersting, K.\n\n\n \n\n\n\n AI and Ethics. August 2024.\n \n\n\n\n
\n\n\n\n \n \n $\"Auditing$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{friedrichAuditingInstructingTexttoimage2024,\n\ttitle = {Auditing and instructing text-to-image generation models on fairness},\n\tissn = {2730-5953, 2730-5961},\n\turl = {https://link.springer.com/10.1007/s43681-024-00531-5},\n\tdoi = {10.1007/s43681-024-00531-5},\n\tabstract = {Abstract \n             \n              Generative AI models have recently achieved astonishing results in quality and are consequently employed in a fast-growing number of applications. However, since they are highly data-driven, relying on billion-sized datasets randomly scraped from the internet, they also suffer from degenerated and biased human behavior, as we demonstrate. In fact, they may even reinforce such biases. To not only uncover but also combat these undesired effects, we present a novel strategy, called \n              Fair Diffusion \n              , to attenuate biases during the deployment of generative text-to-image models. Specifically, we demonstrate shifting a bias in any direction based on human instructions yielding arbitrary proportions for, e.g., identity groups. As our empirical evaluation demonstrates, this introduced control enables instructing generative image models on fairness, requiring no data filtering nor additional training.},\n\tlanguage = {en},\n\turldate = {2025-05-13},\n\tjournal = {AI and Ethics},\n\tauthor = {Friedrich, Felix and Brack, Manuel and Struppek, Lukas and Hintersdorf, Dominik and Schramowski, Patrick and Luccioni, Sasha and Kersting, Kristian},\n\tmonth = aug,\n\tyear = {2024},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Does CLIP Know My Face?.\n \n \n \n \n\n\n \n Hintersdorf, D.; Struppek, L.; Brack, M.; Friedrich, F.; Schramowski, P.; and Kersting, K.\n\n\n \n\n\n\n Journal of Artificial Intelligence Research, 80: 1033–1062. July 2024.\n \n\n\n\n
\n\n\n\n \n \n $\"Does$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{hintersdorfDoesCLIPKnow2024,\n\ttitle = {Does {CLIP} {Know} {My} {Face}?},\n\tvolume = {80},\n\tissn = {1076-9757},\n\turl = {https://www.jair.org/index.php/jair/article/view/15461},\n\tdoi = {10.1613/jair.1.15461},\n\tabstract = {With the rise of deep learning in various applications, privacy concerns around the protection of training data have become a critical area of research. Whereas prior studies have focused on privacy risks in single-modal models, we introduce a novel method to assess privacy for multi-modal models, specifically vision-language models like CLIP. The proposed Identity Inference Attack (IDIA) reveals whether an individual was included in the training data by querying the model with images of the same person. Letting the model choose from a wide variety of possible text labels, the model reveals whether it recognizes the person and, therefore, was used for training. Our large-scale experiments on CLIP demonstrate that individuals used for training can be identified with very high accuracy. We confirm that the model has learned to associate names with depicted individuals, implying the existence of sensitive information that can be extracted by adversaries. Our results highlight the need for stronger privacy protection in large-scale models and suggest that IDIAs can be used to prove the unauthorized use of data for training and to enforce privacy laws. \nThis article appears in the AI \\& Society track.},\n\turldate = {2025-05-13},\n\tjournal = {Journal of Artificial Intelligence Research},\n\tauthor = {Hintersdorf, Dominik and Struppek, Lukas and Brack, Manuel and Friedrich, Felix and Schramowski, Patrick and Kersting, Kristian},\n\tmonth = jul,\n\tyear = {2024},\n\tpages = {1033--1062},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Ledits++: Limitless image editing using text-to-image models.\n \n \n \n\n\n \n Brack, M.; Friedrich, F.; Kornmeier, K.; Tsaban, L.; Schramowski, P.; Kersting, K.; and Passos, A.\n\n\n \n\n\n\n In Proceedings of the IEEE/CVF, pages 8861–8870, 2024. \n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{brackLeditsLimitlessImage2024,\n\ttitle = {Ledits++: {Limitless} image editing using text-to-image models},\n\tbooktitle = {Proceedings of the {IEEE}/{CVF}},\n\tauthor = {Brack, Manuel and Friedrich, Felix and Kornmeier, Katharina and Tsaban, Linoy and Schramowski, Patrick and Kersting, Kristian and Passos, Apolinário},\n\tyear = {2024},\n\tpages = {8861--8870},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n LLMs Lost in Translation: M-ALERT uncovers Cross-Linguistic Safety Gaps.\n \n \n \n \n\n\n \n Friedrich, F.; Tedeschi, S.; Schramowski, P.; Brack, M.; Navigli, R.; Nguyen, H.; Li, B.; and Kersting, K.\n\n\n \n\n\n\n 2024.\n \n\n\n\n
\n\n\n\n \n \n $\"LLMs$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n\n\n\n

@misc{friedrichLLMsLostTranslation2024,\n\ttitle = {{LLMs} {Lost} in {Translation}: {M}-{ALERT} uncovers {Cross}-{Linguistic} {Safety} {Gaps}},\n\tcopyright = {Creative Commons Attribution 4.0 International},\n\tshorttitle = {{LLMs} {Lost} in {Translation}},\n\turl = {https://arxiv.org/abs/2412.15035},\n\tdoi = {10.48550/ARXIV.2412.15035},\n\tabstract = {Building safe Large Language Models (LLMs) across multiple languages is essential in ensuring both safe access and linguistic diversity. To this end, we introduce M-ALERT, a multilingual benchmark that evaluates the safety of LLMs in five languages: English, French, German, Italian, and Spanish. M-ALERT includes 15k high-quality prompts per language, totaling 75k, following the detailed ALERT taxonomy. Our extensive experiments on 10 state-of-the-art LLMs highlight the importance of language-specific safety analysis, revealing that models often exhibit significant inconsistencies in safety across languages and categories. For instance, Llama3.2 shows high unsafety in the category crime\\_tax for Italian but remains safe in other languages. Similar differences can be observed across all models. In contrast, certain categories, such as substance\\_cannabis and crime\\_propaganda, consistently trigger unsafe responses across models and languages. These findings underscore the need for robust multilingual safety practices in LLMs to ensure safe and responsible usage across diverse user communities.},\n\turldate = {2025-05-13},\n\tpublisher = {arXiv},\n\tauthor = {Friedrich, Felix and Tedeschi, Simone and Schramowski, Patrick and Brack, Manuel and Navigli, Roberto and Nguyen, Huu and Li, Bo and Kersting, Kristian},\n\tyear = {2024},\n\tkeywords = {Computation and Language (cs.CL), FOS: Computer and information sciences},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n DeiSAM: Segment Anything with Deictic Prompting.\n \n \n \n\n\n \n Shindo, H.; Brack, M.; Sudhakaran, G.; Dhami, D. S.; Schramowski, P.; and Kersting, K.\n\n\n \n\n\n\n In volume 37, pages 52266–52295, 2024. \n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{shindoDeiSAMSegmentAnything2024,\n\ttitle = {{DeiSAM}: {Segment} {Anything} with {Deictic} {Prompting}},\n\tvolume = {37},\n\tauthor = {Shindo, Hikaru and Brack, Manuel and Sudhakaran, Gopika and Dhami, Devendra Singh and Schramowski, Patrick and Kersting, Kristian},\n\tyear = {2024},\n\tpages = {52266--52295},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n SCAR: Sparse Conditioned Autoencoders for Concept Detection and Steering in LLMs.\n \n \n \n \n\n\n \n Härle, R.; Friedrich, F.; Brack, M.; Deiseroth, B.; Schramowski, P.; and Kersting, K.\n\n\n \n\n\n\n 2024.\n \n\n\n\n
\n\n\n\n \n \n $\"SCAR:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n\n\n\n

@misc{harleSCARSparseConditioned2024,\n\ttitle = {{SCAR}: {Sparse} {Conditioned} {Autoencoders} for {Concept} {Detection} and {Steering} in {LLMs}},\n\tcopyright = {Creative Commons Attribution Non Commercial Share Alike 4.0 International},\n\tshorttitle = {{SCAR}},\n\turl = {https://arxiv.org/abs/2411.07122},\n\tdoi = {10.48550/ARXIV.2411.07122},\n\tabstract = {Large Language Models (LLMs) have demonstrated remarkable capabilities in generating human-like text, but their output may not be aligned with the user or even produce harmful content. This paper presents a novel approach to detect and steer concepts such as toxicity before generation. We introduce the Sparse Conditioned Autoencoder (SCAR), a single trained module that extends the otherwise untouched LLM. SCAR ensures full steerability, towards and away from concepts (e.g., toxic content), without compromising the quality of the model's text generation on standard evaluation benchmarks. We demonstrate the effective application of our approach through a variety of concepts, including toxicity, safety, and writing style alignment. As such, this work establishes a robust framework for controlling LLM generations, ensuring their ethical and safe deployment in real-world applications.},\n\turldate = {2025-05-13},\n\tpublisher = {arXiv},\n\tauthor = {Härle, Ruben and Friedrich, Felix and Brack, Manuel and Deiseroth, Björn and Schramowski, Patrick and Kersting, Kristian},\n\tyear = {2024},\n\tkeywords = {Computation and Language (cs.CL), FOS: Computer and information sciences},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Human-AI Engineering for Adults.\n \n \n \n \n\n\n \n Meyer-Vitali, A.; and Mulder, W.\n\n\n \n\n\n\n In HHAI 2024: Hybrid Human AI Systems for the Social Good, pages 228–240. IOS Press, 2024.\n \n\n\n\n
\n\n\n\n \n \n $\"Human-AI$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@incollection{meyer-vitaliHumanAIEngineeringAdults2024,\n\ttitle = {Human-{AI} {Engineering} for {Adults}},\n\tcopyright = {All rights reserved},\n\turl = {https://ebooks.iospress.nl/doi/10.3233/FAIA240197},\n\tdoi = {10.3233/FAIA240197},\n\turldate = {2024-06-08},\n\tbooktitle = {{HHAI} 2024: {Hybrid} {Human} {AI} {Systems} for the {Social} {Good}},\n\tpublisher = {IOS Press},\n\tauthor = {Meyer-Vitali, Andr\\&\\#233 and Mulder, Wico},\n\tyear = {2024},\n\tpages = {228--240},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Engineering Principles for Building Trusted Human-AI Systems.\n \n \n \n\n\n \n Meyer-Vitali, A.; and Mulder, W.\n\n\n \n\n\n\n In Arai, K., editor(s), Intelligent Systems and Applications, pages 468–485, Cham, 2024. Springer Nature Switzerland\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{meyer-vitaliEngineeringPrinciplesBuilding2024,\n\taddress = {Cham},\n\ttitle = {Engineering {Principles} for {Building} {Trusted} {Human}-{AI} {Systems}},\n\tcopyright = {All rights reserved},\n\tisbn = {978-3-031-66428-1},\n\tdoi = {10.1007/978-3-031-66428-1_30},\n\tabstract = {In the process engineering reliable and trustworthy AI systems there is significant wisdom to be gained from traditional engineering domains. Extending on earlier work our attention is on topics that stress the principles of building human-AI systems. We plea for a reinforced attention for engineering methods and processes in order to urge the essence for improved scientific progress and industrial AI applications where one can stand on the shoulders of giants. On the one hand, we see their complexity increase on an individual level, as well as on their connected dependency levels, whilst on the other hand, we see a growing lack of experience on the level of their design and engineering. The complexity of current AI models often limits our understanding. The methods and processes to ensure safety, reliability, and transparency are insufficient. This poses serious risks at the level of trustworthiness, particularly when it comes to critical applications with significant social, economic or even physical impact. Future AI systems must adhere to stringent requirements, as mandated, for instance, by the European AI Act, ensuring meticulous design, validation, and certification based on clearly defined criteria.},\n\tlanguage = {en},\n\tbooktitle = {Intelligent {Systems} and {Applications}},\n\tpublisher = {Springer Nature Switzerland},\n\tauthor = {Meyer-Vitali, André and Mulder, Wico},\n\teditor = {Arai, Kohei},\n\tyear = {2024},\n\tpages = {468--485},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n ALERT: A Comprehensive Benchmark for Assessing Large Language Models' Safety through Red Teaming.\n \n \n \n \n\n\n \n Tedeschi, S.; Friedrich, F.; Schramowski, P.; Kersting, K.; Navigli, R.; Nguyen, H.; and Li, B.\n\n\n \n\n\n\n June 2024.\n arXiv:2404.08676 [cs]\n\n\n\n
\n\n\n\n \n \n $\"ALERT:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n\n\n\n

@misc{tedeschiALERTComprehensiveBenchmark2024,\n\ttitle = {{ALERT}: {A} {Comprehensive} {Benchmark} for {Assessing} {Large} {Language} {Models}' {Safety} through {Red} {Teaming}},\n\tshorttitle = {{ALERT}},\n\turl = {http://arxiv.org/abs/2404.08676},\n\tdoi = {10.48550/arXiv.2404.08676},\n\tabstract = {When building Large Language Models (LLMs), it is paramount to bear safety in mind and protect them with guardrails. Indeed, LLMs should never generate content promoting or normalizing harmful, illegal, or unethical behavior that may contribute to harm to individuals or society. This principle applies to both normal and adversarial use. In response, we introduce ALERT, a large-scale benchmark to assess safety based on a novel fine-grained risk taxonomy. It is designed to evaluate the safety of LLMs through red teaming methodologies and consists of more than 45k instructions categorized using our novel taxonomy. By subjecting LLMs to adversarial testing scenarios, ALERT aims to identify vulnerabilities, inform improvements, and enhance the overall safety of the language models. Furthermore, the fine-grained taxonomy enables researchers to perform an in-depth evaluation that also helps one to assess the alignment with various policies. In our experiments, we extensively evaluate 10 popular open- and closed-source LLMs and demonstrate that many of them still struggle to attain reasonable levels of safety.},\n\turldate = {2024-12-04},\n\tpublisher = {arXiv},\n\tauthor = {Tedeschi, Simone and Friedrich, Felix and Schramowski, Patrick and Kersting, Kristian and Navigli, Roberto and Nguyen, Huu and Li, Bo},\n\tmonth = jun,\n\tyear = {2024},\n\tnote = {arXiv:2404.08676 [cs]},\n\tkeywords = {Computer Science - Computation and Language, Computer Science - Computers and Society, Computer Science - Machine Learning},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Modular Design Patterns for Generative Neuro-Symbolic Systems.\n \n \n \n \n\n\n \n Boer, M. H. T. d.; Smit, Q. S.; Bekkum, M. v.; Meyer-Vitali, A.; and Schmid, T.\n\n\n \n\n\n\n In Sartini, B.; Raad, J.; Lisena, P.; Peñuela, A. M.; Beetz, M.; Blin, I.; Cimiano, P.; Berardinis, J. d.; Gottschalk, S.; Ilievski, F.; Jain, N.; Kim, J.; Kümpel, M.; Motta, E.; Tiddi, I.; and Töberg, J., editor(s), Joint Proceedings of the ESWC 2024 Workshops and Tutorials, volume 3749, of CEUR Workshop Proceedings, Hersonissos, Greece, May 2024. CEUR\n \n\n\n\n
\n\n\n\n \n \n $\"Modular$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{boerModularDesignPatterns2024,\n\taddress = {Hersonissos, Greece},\n\tseries = {{CEUR} {Workshop} {Proceedings}},\n\ttitle = {Modular {Design} {Patterns} for {Generative} {Neuro}-{Symbolic} {Systems}},\n\tvolume = {3749},\n\tissn = {1613-0073},\n\turl = {https://ceur-ws.org/Vol-3749/#genesy-03},\n\tlanguage = {en},\n\turldate = {2024-09-11},\n\tbooktitle = {Joint {Proceedings} of the {ESWC} 2024 {Workshops} and {Tutorials}},\n\tpublisher = {CEUR},\n\tauthor = {Boer, Maaike H. T. de and Smit, Quirine S. and Bekkum, Michael van and Meyer-Vitali, André and Schmid, Thomas},\n\teditor = {Sartini, Bruno and Raad, Joe and Lisena, Pasquale and Peñuela, Albert Meroño and Beetz, Michael and Blin, Inès and Cimiano, Philipp and Berardinis, Jacopo de and Gottschalk, Simon and Ilievski, Filip and Jain, Nitisha and Kim, Jongmo and Kümpel, Michaela and Motta, Enrico and Tiddi, Ilaria and Töberg, Jan-Philipp},\n\tmonth = may,\n\tyear = {2024},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Trusted AI – Vertrauenswürdigkeit und große Sprachmodelle.\n \n \n \n \n\n\n \n Meyer-Vitali, A.; and Ostermann, S.\n\n\n \n\n\n\n dfki ai next, 2024(2): 6–7. September 2024.\n \n\n\n\n
\n\n\n\n \n \n $\"Trusted$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{meyer-vitaliTrustedAIVertrauenswuerdigkeit2024,\n\ttitle = {Trusted {AI} – {Vertrauenswürdigkeit} und große  {Sprachmodelle}},\n\tvolume = {2024},\n\turl = {https://www.dfki.de/web/news-media/news/dfki-ai-next},\n\tlanguage = {German},\n\tnumber = {2},\n\tjournal = {dfki ai next},\n\tauthor = {Meyer-Vitali, André and Ostermann, Simon},\n\tmonth = sep,\n\tyear = {2024},\n\tpages = {6--7},\n}\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n

\n\n\n\n

\n\n\n\n\n\n

\n\n

\n \n 2023\n \n \n (22)\n \n \n

\n \n \n

\n \n\n \n \n \n \n \n \n Investigating the Encoding of Words in BERT`s Neurons Using Feature Textualization.\n \n \n \n \n\n\n \n Baeumel, T.; Vijayakumar, S.; van Genabith, J.; Neumann, G.; and Ostermann, S.\n\n\n \n\n\n\n In Belinkov, Y.; Hao, S.; Jumelet, J.; Kim, N.; McCarthy, A.; and Mohebbi, H., editor(s), Proceedings of the 6th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP, pages 261–270, Singapore, December 2023. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"Investigating$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{baeumelInvestigatingEncodingWords2023,\n\taddress = {Singapore},\n\ttitle = {Investigating the {Encoding} of {Words} in {BERT}`s {Neurons} {Using} {Feature} {Textualization}},\n\turl = {https://aclanthology.org/2023.blackboxnlp-1.20/},\n\tdoi = {10.18653/v1/2023.blackboxnlp-1.20},\n\tabstract = {Pretrained language models (PLMs) form the basis of most state-of-the-art NLP technologies. Nevertheless, they are essentially black boxes: Humans do not have a clear understanding of what knowledge is encoded in different parts of the models, especially in individual neurons. A contrast is in computer vision, where feature visualization provides a decompositional interpretability technique for neurons of vision models. Activation maximization is used to synthesize inherently interpretable visual representations of the information encoded in individual neurons. Our work is inspired by this but presents a cautionary tale on the interpretability of single neurons, based on the first large-scale attempt to adapt activation maximization to NLP, and, more specifically, large PLMs. We propose feature textualization, a technique to produce dense representations of neurons in the PLM word embedding space. We apply feature textualization to the BERT model to investigate whether the knowledge encoded in individual neurons can be interpreted and symbolized. We find that the produced representations can provide insights about the knowledge encoded in individual neurons, but that individual neurons do not represent clear-cut symbolic units of language such as words. Additionally, we use feature textualization to investigate how many neurons are needed to encode words in BERT.},\n\turldate = {2025-05-22},\n\tbooktitle = {Proceedings of the 6th {BlackboxNLP} {Workshop}: {Analyzing} and {Interpreting} {Neural} {Networks} for {NLP}},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Baeumel, Tanja and Vijayakumar, Soniya and van Genabith, Josef and Neumann, Guenter and Ostermann, Simon},\n\teditor = {Belinkov, Yonatan and Hao, Sophie and Jumelet, Jaap and Kim, Najoung and McCarthy, Arya and Mohebbi, Hosein},\n\tmonth = dec,\n\tyear = {2023},\n\tpages = {261--270},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n XAI Requirements in Smart Production Processes: A Case Study.\n \n \n \n\n\n \n Baum, D.; Baum, K.; Gros, T. P.; and Wolf, V.\n\n\n \n\n\n\n In Longo, L., editor(s), Explainable Artificial Intelligence. Proceedings of the World Conference on eXplainable Artificial Intelligence (xAI 2023), volume 1901, of Communications in Computer and Information Science (CCIS), pages 3–24, Cham, 2023. Springer Nature Switzerland\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{baumXAIRequirementsSmart2023,\n\taddress = {Cham},\n\tseries = {Communications in {Computer} and {Information} {Science} ({CCIS})},\n\ttitle = {{XAI} {Requirements} in {Smart} {Production} {Processes}: {A} {Case} {Study}},\n\tvolume = {1901},\n\tdoi = {10.1007/978-3-031-44064-9_1},\n\tabstract = {The increasing prevalence of artificial intelligence (AI) systems has led to a growing consensus on the importance of the explainability of such systems. This is often emphasized with respect to societal and developmental contexts, but it is also crucial within the context of business processes, including manufacturing and production. While this is widely recognized, there is a notable lack of practical examples that demonstrate how to take explainability into account in the latter contexts. This paper presents a real-world use case in which we employed AI to optimize an Industry 4.0 production process without considering explainable AI (XAI) requirements. Building on previous work on models of the relationship between XAI methods and various associated expectations, as well as non-functional explainability requirements, we show how business-oriented XAI requirements can be formulated and prepared for integration into process design. This case study is a valuable resource for researchers and practitioners seeking better to understand the role of explainable AI in practice.},\n\tbooktitle = {Explainable {Artificial} {Intelligence}. {Proceedings} of the {World} {Conference} on {eXplainable} {Artificial} {Intelligence} ({xAI} 2023)},\n\tpublisher = {Springer Nature Switzerland},\n\tauthor = {Baum, Deborah and Baum, Kevin and Gros, Timo P. and Wolf, Verena},\n\teditor = {Longo, Luca},\n\tyear = {2023},\n\tpages = {3--24},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n From Fear to Action: AI Governance and Opportunities for All.\n \n \n \n\n\n \n Baum, K.; Bryson, J.; Dignum, F.; Dignum, V.; Grobelnik, M.; Hoos, H.; Irgens, M.; Lukowicz, P.; Muller, C.; Rossi, F.; Shawe-Taylor, J.; Theodorou, A.; and Vinuesa, R.\n\n\n \n\n\n\n Frontiers in Computer Science, 5. 2023.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{baumFearActionAI2023,\n\ttitle = {From {Fear} to {Action}: {AI} {Governance} and {Opportunities} for {All}},\n\tvolume = {5},\n\tdoi = {10.3389/fcomp.2023.1210421},\n\tjournal = {Frontiers in Computer Science},\n\tauthor = {Baum, Kevin and Bryson, Joanna and Dignum, Frank and Dignum, Virginia and Grobelnik, Marko and Hoos, Holger and Irgens, Morten and Lukowicz, Paul and Muller, Catelijne and Rossi, Francesca and Shawe-Taylor, John and Theodorou, Andreas and Vinuesa, Ricardo},\n\tyear = {2023},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Increasing effect sizes of pairwise conditional independence tests between random vectors.\n \n \n \n\n\n \n Hochsprung, T.; Wahl, J.; Gerhardus, A.; Ninad, U.; and Runge, J.\n\n\n \n\n\n\n In Uncertainty in Artificial Intelligence, pages 879–889, 2023. PMLR\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{hochsprungIncreasingEffectSizes2023,\n\ttitle = {Increasing effect sizes of pairwise conditional independence tests between random vectors},\n\tbooktitle = {Uncertainty in {Artificial} {Intelligence}},\n\tpublisher = {PMLR},\n\tauthor = {Hochsprung, Tom and Wahl, Jonas and Gerhardus, Andreas and Ninad, Urmi and Runge, Jakob},\n\tyear = {2023},\n\tpages = {879--889},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Vector causal inference between two groups of variables.\n \n \n \n\n\n \n Wahl, J.; Ninad, U.; and Runge, J.\n\n\n \n\n\n\n In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, pages 12305–12312, 2023. \n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{wahlVectorCausalInference2023,\n\ttitle = {Vector causal inference between two groups of variables},\n\tvolume = {37},\n\tbooktitle = {Proceedings of the {AAAI} {Conference} on {Artificial} {Intelligence}},\n\tauthor = {Wahl, Jonas and Ninad, Urmi and Runge, Jakob},\n\tyear = {2023},\n\tpages = {12305--12312},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Find-2-Find: Multitask Learning for Anaphora Resolution and Object Localization.\n \n \n \n \n\n\n \n Oguz, C.; Denis, P.; Vincent, E.; Ostermann, S.; and van Genabith, J.\n\n\n \n\n\n\n In Bouamor, H.; Pino, J.; and Bali, K., editor(s), Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 8099–8110, Singapore, December 2023. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"Find-2-Find:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n \n \n 1 download\n \n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{oguzFind2FindMultitaskLearning2023,\n\taddress = {Singapore},\n\ttitle = {Find-2-{Find}: {Multitask} {Learning} for {Anaphora} {Resolution} and {Object} {Localization}},\n\tshorttitle = {Find-2-{Find}},\n\turl = {https://aclanthology.org/2023.emnlp-main.504/},\n\tdoi = {10.18653/v1/2023.emnlp-main.504},\n\tabstract = {In multimodal understanding tasks, visual and linguistic ambiguities can arise. Visual ambiguity can occur when visual objects require a model to ground a referring expression in a video without strong supervision, while linguistic ambiguity can occur from changes in entities in action flows. As an example from the cooking domain, “oil” mixed with “salt” and “pepper” could later be referred to as a “mixture”. Without a clear visual-linguistic alignment, we cannot know which among several objects shown is referred to by the language expression “mixture”, and without resolved antecedents, we cannot pinpoint what the mixture is. We define this chicken-and-egg problem as Visual-linguistic Ambiguity. In this paper, we present Find2Find, a joint anaphora resolution and object localization dataset targeting the problem of visual-linguistic ambiguity, consisting of 500 anaphora-annotated recipes with corresponding videos. We present experimental results of a novel end-to-end joint multitask learning framework for Find2Find that fuses visual and textual information and shows improvements both for anaphora resolution and object localization with one joint model in multitask learning, as compared to a strong single-task baseline.},\n\turldate = {2025-05-22},\n\tbooktitle = {Proceedings of the 2023 {Conference} on {Empirical} {Methods} in {Natural} {Language} {Processing}},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Oguz, Cennet and Denis, Pascal and Vincent, Emmanuel and Ostermann, Simon and van Genabith, Josef},\n\teditor = {Bouamor, Houda and Pino, Juan and Bali, Kalika},\n\tmonth = dec,\n\tyear = {2023},\n\tpages = {8099--8110},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Speaking Multiple Languages Affects the Moral Bias of Language Models.\n \n \n \n \n\n\n \n Haemmerl, K.; Deiseroth, B.; Schramowski, P.; Libovický, J.; Rothkopf, C.; Fraser, A.; and Kersting, K.\n\n\n \n\n\n\n In Findings of the Association for Computational Linguistics: ACL 2023, pages 2137–2156, Toronto, Canada, 2023. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n \n $\"Speaking$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{haemmerlSpeakingMultipleLanguages2023,\n\taddress = {Toronto, Canada},\n\ttitle = {Speaking {Multiple} {Languages} {Affects} the {Moral} {Bias} of {Language} {Models}},\n\turl = {https://aclanthology.org/2023.findings-acl.134},\n\tdoi = {10.18653/v1/2023.findings-acl.134},\n\tlanguage = {en},\n\turldate = {2025-05-20},\n\tbooktitle = {Findings of the {Association} for {Computational} {Linguistics}: {ACL} 2023},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Haemmerl, Katharina and Deiseroth, Bjoern and Schramowski, Patrick and Libovický, Jindřich and Rothkopf, Constantin and Fraser, Alexander and Kersting, Kristian},\n\tyear = {2023},\n\tpages = {2137--2156},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n A typology for exploring the mitigation of shortcut behaviour.\n \n \n \n \n\n\n \n Friedrich, F.; Stammer, W.; Schramowski, P.; and Kersting, K.\n\n\n \n\n\n\n Nature Machine Intelligence, 5(3): 319–330. March 2023.\n \n\n\n\n
\n\n\n\n \n \n $\"A$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{friedrichTypologyExploringMitigation2023a,\n\ttitle = {A typology for exploring the mitigation of shortcut behaviour},\n\tvolume = {5},\n\tissn = {2522-5839},\n\turl = {https://www.nature.com/articles/s42256-023-00612-w},\n\tdoi = {10.1038/s42256-023-00612-w},\n\tlanguage = {en},\n\tnumber = {3},\n\turldate = {2025-05-21},\n\tjournal = {Nature Machine Intelligence},\n\tauthor = {Friedrich, Felix and Stammer, Wolfgang and Schramowski, Patrick and Kersting, Kristian},\n\tmonth = mar,\n\tyear = {2023},\n\tpages = {319--330},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n SEGA: Instructing Text-to-Image Models using Semantic Guidance.\n \n \n \n \n\n\n \n Brack, M.; Friedrich, F.; Hintersdorf, D.; Struppek, L.; Schramowski, P.; and Kersting, K.\n\n\n \n\n\n\n In In Proceedings of the 37th Conference on Neural Information Processing Systems, 2023. arXiv\n \n\n\n\n
\n\n\n\n \n \n $\"SEGA:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n \n \n\n\n\n

@inproceedings{brackSEGAInstructingTexttoImage2023,\n\ttitle = {{SEGA}: {Instructing} {Text}-to-{Image} {Models} using {Semantic} {Guidance}},\n\tcopyright = {Creative Commons Attribution 4.0 International},\n\tshorttitle = {{SEGA}},\n\turl = {https://arxiv.org/abs/2301.12247},\n\tdoi = {10.48550/ARXIV.2301.12247},\n\tabstract = {Text-to-image diffusion models have recently received a lot of interest for their astonishing ability to produce high-fidelity images from text only. However, achieving one-shot generation that aligns with the user's intent is nearly impossible, yet small changes to the input prompt often result in very different images. This leaves the user with little semantic control. To put the user in control, we show how to interact with the diffusion process to flexibly steer it along semantic directions. This semantic guidance (SEGA) generalizes to any generative architecture using classifier-free guidance. More importantly, it allows for subtle and extensive edits, changes in composition and style, as well as optimizing the overall artistic conception. We demonstrate SEGA's effectiveness on both latent and pixel-based diffusion models such as Stable Diffusion, Paella, and DeepFloyd-IF using a variety of tasks, thus providing strong evidence for its versatility, flexibility, and improvements over existing methods.},\n\turldate = {2025-05-21},\n\tbooktitle = {In {Proceedings} of the 37th {Conference} on {Neural} {Information} {Processing} {Systems}},\n\tpublisher = {arXiv},\n\tauthor = {Brack, Manuel and Friedrich, Felix and Hintersdorf, Dominik and Struppek, Lukas and Schramowski, Patrick and Kersting, Kristian},\n\tyear = {2023},\n\tkeywords = {Artificial Intelligence (cs.AI), Computer Vision and Pattern Recognition (cs.CV), FOS: Computer and information sciences, Machine Learning (cs.LG)},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Hyperspectral Imaging in the UV Range Allows for Differentiation of Sugar Beet Diseases Based on Changes in Secondary Plant Metabolites.\n \n \n \n \n\n\n \n Brugger, A.; Yamati, F. I.; Barreto, A.; Paulus, S.; Schramowsk, P.; Kersting, K.; Steiner, U.; Neugart, S.; and Mahlein, A.\n\n\n \n\n\n\n Phytopathology®, 113(1): 44–54. January 2023.\n \n\n\n\n
\n\n\n\n \n \n $\"Hyperspectral$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{bruggerHyperspectralImagingUV2023a,\n\ttitle = {Hyperspectral {Imaging} in the {UV} {Range} {Allows} for {Differentiation} of {Sugar} {Beet} {Diseases} {Based} on {Changes} in {Secondary} {Plant} {Metabolites}},\n\tvolume = {113},\n\tissn = {0031-949X, 1943-7684},\n\turl = {https://apsjournals.apsnet.org/doi/10.1094/PHYTO-03-22-0086-R},\n\tdoi = {10.1094/PHYTO-03-22-0086-R},\n\tabstract = {Fungal infections trigger defense or signaling responses in plants, leading to various changes in plant metabolites. The changes in metabolites, for example chlorophyll or flavonoids, have long been detectable using time-consuming destructive analytical methods including high-performance liquid chromatography or photometric determination. Recent plant phenotyping studies have revealed that hyperspectral imaging (HSI) in the UV range can be used to link spectral changes with changes in plant metabolites. To compare established destructive analytical methods with new nondestructive hyperspectral measurements, the interaction between sugar beet leaves and the pathogens Cercospora beticola, which causes Cercospora leaf spot disease (CLS), and Uromyces betae, which causes sugar beet rust (BR), was investigated. With the help of destructive analyses, we showed that both diseases have different effects on chlorophylls, carotenoids, flavonoids, and several phenols. Nondestructive hyperspectral measurements in the UV range revealed different effects of CLS and BR on plant metabolites resulting in distinct reflectance patterns. Both diseases resulted in specific spectral changes that allowed differentiation between the two diseases. Machine learning algorithms enabled the differentiation between the symptom classes and recognition of the two sugar beet diseases. Feature importance analysis identified specific wavelengths important to the classification, highlighting the utility of the UV range. The study demonstrates that HSI in the UV range is a promising, nondestructive tool to investigate the influence of plant diseases on plant physiology and biochemistry.},\n\tlanguage = {en},\n\tnumber = {1},\n\turldate = {2025-05-21},\n\tjournal = {Phytopathology®},\n\tauthor = {Brugger, Anna and Yamati, Facundo Ispizua and Barreto, Abel and Paulus, Stefan and Schramowsk, Patrick and Kersting, Kristian and Steiner, Ulrike and Neugart, Susanne and Mahlein, Anne-Katrin},\n\tmonth = jan,\n\tyear = {2023},\n\tpages = {44--54},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n MultiFusion: Fusing Pre-Trained Models for Multi-Lingual, Multi-Modal Image Generation.\n \n \n \n \n\n\n \n Bellagente, M.; Brack, M.; Teufel, H.; Friedrich, F.; Deiseroth, B.; Eichenberg, C.; Dai, A.; Baldock, R.; Nanda, S.; Oostermeijer, K.; Cruz-Salinas, A. F.; Schramowski, P.; Kersting, K.; and Weinbach, S.\n\n\n \n\n\n\n In Proceedings of the 37th Conference on Neural Information Processing Systems, 2023. arXiv\n \n\n\n\n
\n\n\n\n \n \n $\"MultiFusion:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n \n \n\n\n\n

@inproceedings{bellagenteMultiFusionFusingPreTrained2023,\n\ttitle = {{MultiFusion}: {Fusing} {Pre}-{Trained} {Models} for {Multi}-{Lingual}, {Multi}-{Modal} {Image} {Generation}},\n\tcopyright = {Creative Commons Attribution 4.0 International},\n\tshorttitle = {{MultiFusion}},\n\turl = {https://arxiv.org/abs/2305.15296},\n\tdoi = {10.48550/ARXIV.2305.15296},\n\tabstract = {The recent popularity of text-to-image diffusion models (DM) can largely be attributed to the intuitive interface they provide to users. The intended generation can be expressed in natural language, with the model producing faithful interpretations of text prompts. However, expressing complex or nuanced ideas in text alone can be difficult. To ease image generation, we propose MultiFusion that allows one to express complex and nuanced concepts with arbitrarily interleaved inputs of multiple modalities and languages. MutliFusion leverages pre-trained models and aligns them for integration into a cohesive system, thereby avoiding the need for extensive training from scratch. Our experimental results demonstrate the efficient transfer of capabilities from individual modules to the downstream model. Specifically, the fusion of all independent components allows the image generation module to utilize multilingual, interleaved multimodal inputs despite being trained solely on monomodal data in a single language.},\n\turldate = {2025-05-21},\n\tbooktitle = {Proceedings of the 37th {Conference} on {Neural} {Information} {Processing} {Systems}},\n\tpublisher = {arXiv},\n\tauthor = {Bellagente, Marco and Brack, Manuel and Teufel, Hannah and Friedrich, Felix and Deiseroth, Björn and Eichenberg, Constantin and Dai, Andrew and Baldock, Robert and Nanda, Souradeep and Oostermeijer, Koen and Cruz-Salinas, Andres Felipe and Schramowski, Patrick and Kersting, Kristian and Weinbach, Samuel},\n\tyear = {2023},\n\tkeywords = {Artificial Intelligence (cs.AI), Computer Vision and Pattern Recognition (cs.CV), FOS: Computer and information sciences, Machine Learning (cs.LG)},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n AtMan: Understanding Transformer Predictions Through Memory Efficient Attention Manipulation.\n \n \n \n \n\n\n \n Deiseroth, B.; Deb, M.; Weinbach, S.; Brack, M.; Schramowski, P.; and Kersting, K.\n\n\n \n\n\n\n In Proceedings of the 37th Conference on Neural Information Processing Systems, 2023. \n \n\n\n\n
\n\n\n\n \n \n $\"AtMan:$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n\n\n\n

@inproceedings{deiserothAtManUnderstandingTransformer2023a,\n\ttitle = {{AtMan}: {Understanding} {Transformer} {Predictions} {Through} {Memory} {Efficient} {Attention} {Manipulation}},\n\tcopyright = {Creative Commons Attribution 4.0 International},\n\tshorttitle = {{AtMan}},\n\turl = {https://arxiv.org/abs/2301.08110},\n\tdoi = {10.48550/ARXIV.2301.08110},\n\tabstract = {Generative transformer models have become increasingly complex, with large numbers of parameters and the ability to process multiple input modalities. Current methods for explaining their predictions are resource-intensive. Most crucially, they require prohibitively large amounts of extra memory, since they rely on backpropagation which allocates almost twice as much GPU memory as the forward pass. This makes it difficult, if not impossible, to use them in production. We present AtMan that provides explanations of generative transformer models at almost no extra cost. Specifically, AtMan is a modality-agnostic perturbation method that manipulates the attention mechanisms of transformers to produce relevance maps for the input with respect to the output prediction. Instead of using backpropagation, AtMan applies a parallelizable token-based search method based on cosine similarity neighborhood in the embedding space. Our exhaustive experiments on text and image-text benchmarks demonstrate that AtMan outperforms current state-of-the-art gradient-based methods on several metrics while being computationally efficient. As such, AtMan is suitable for use in large model inference deployments.},\n\turldate = {2025-05-21},\n\tbooktitle = {Proceedings of the 37th {Conference} on {Neural} {Information} {Processing} {Systems}},\n\tauthor = {Deiseroth, Björn and Deb, Mayukh and Weinbach, Samuel and Brack, Manuel and Schramowski, Patrick and Kersting, Kristian},\n\tyear = {2023},\n\tkeywords = {Artificial Intelligence (cs.AI), FOS: Computer and information sciences, Machine Learning (cs.LG)},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Revision Transformers: Instructing Language Models to Change Their Values.\n \n \n \n \n\n\n \n Friedrich, F.; Stammer, W.; Schramowski, P.; and Kersting, K.\n\n\n \n\n\n\n In ECAI 2023, pages 756–763. IOS Press, 2023.\n \n\n\n\n
\n\n\n\n \n \n $\"Revision$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@incollection{friedrichRevisionTransformersInstructing2023a,\n\ttitle = {Revision {Transformers}: {Instructing} {Language} {Models} to {Change} {Their} {Values}},\n\tshorttitle = {Revision {Transformers}},\n\turl = {https://ebooks.iospress.nl/doi/10.3233/FAIA230341},\n\tdoi = {10.3233/FAIA230341},\n\tlanguage = {en},\n\turldate = {2025-05-20},\n\tbooktitle = {{ECAI} 2023},\n\tpublisher = {IOS Press},\n\tauthor = {Friedrich, Felix and Stammer, Wolfgang and Schramowski, Patrick and Kersting, Kristian},\n\tyear = {2023},\n\tpages = {756--763},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n ILLUME: Rationalizing Vision-Language Models through Human Interactions.\n \n \n \n\n\n \n Brack, M.; Schramowski, P.; Deiseroth, B.; and Kersting, K.\n\n\n \n\n\n\n In Proceedings of the 40th International Conference on Machine Learning, volume 202, pages 3021–3037, 2023. \n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{brackILLUMERationalizingVisionLanguage2023,\n\ttitle = {{ILLUME}: {Rationalizing} {Vision}-{Language} {Models} through {Human} {Interactions}},\n\tvolume = {202},\n\tbooktitle = {Proceedings of the 40th {International} {Conference} on {Machine} {Learning}},\n\tauthor = {Brack, Manuel and Schramowski, Patrick and Deiseroth, Björn and Kersting, Kristian},\n\tyear = {2023},\n\tpages = {3021--3037},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Self-Supervised Learning of Machine Ethics.\n \n \n \n \n\n\n \n Schramowski, P.\n\n\n \n\n\n\n . 2023.\n \n\n\n\n
\n\n\n\n \n \n $\"Self-Supervised$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{schramowskiSelfSupervisedLearningMachine2023,\n\ttitle = {Self-{Supervised} {Learning} of {Machine} {Ethics}},\n\tcopyright = {Creative Commons Attribution Share Alike 4.0 International},\n\turl = {https://tuprints.ulb.tu-darmstadt.de/id/eprint/23090},\n\tdoi = {10.26083/TUPRINTS-00023090},\n\tabstract = {In recent years Artificial Intelligence (AI), especially deep learning, has proven to be a technology driver in industry. However, while advancing existing and creating novel technologies, automatizing processes, and assisting humans in essential areas such as drug discovery, they raise many concerns, like other groundbreaking novel technologies before. In this case, these concerns include, for instance, models producing stereotypical and derogatory content as well as gender and racial biases. Since AI technologies will permeate more of our lives in the coming years, these concerns need to be addressed. This thesis examines recent data-driven approaches, which often suffer from degenerated and biased behavior through their self-supervised training on large-scale noisy web data, containing potential inappropriate data. While this is well-established, we will investigate and demonstrate the promises of deep models’ acquired knowledge and capabilities through the provision of this very particular potentially inappropriate data. Importantly, we present the first approaches for learning ethics from data. Our findings suggest that if we build an AI system that learns an improved representation of data and that is able to better understand and produce it, in the process, it will also acquire more accurate societal knowledge, in this case, historical cultural associations to make human-like "right" and "wrong" choices. Furthermore, based on these findings, we consequently ask the arguably "circular" question of whether a machine can help us mitigate their associated concerns. Importantly, we demonstrate the importance of their ability to distinguish between "right" and "wrong" and how utilizing them can mitigate associated risks surrounding large-scale models themselves. However, we also highlight the role of human-machine interaction to explore and reinforce AI systems’ properties, including their flaws and merits, and present how human feedback on explanations can align deep learning based models with our precepts. We present these algorithms and corresponding findings, providing important insights for the goal of putting human values into AI systems, which, summarized, may not be insurmountable in the long run.},\n\turldate = {2025-05-13},\n\tauthor = {Schramowski, Patrick},\n\tyear = {2023},\n}\n\n\n\n

\n\n\n

\n In recent years Artificial Intelligence (AI), especially deep learning, has proven to be a technology driver in industry. However, while advancing existing and creating novel technologies, automatizing processes, and assisting humans in essential areas such as drug discovery, they raise many concerns, like other groundbreaking novel technologies before. In this case, these concerns include, for instance, models producing stereotypical and derogatory content as well as gender and racial biases. Since AI technologies will permeate more of our lives in the coming years, these concerns need to be addressed. This thesis examines recent data-driven approaches, which often suffer from degenerated and biased behavior through their self-supervised training on large-scale noisy web data, containing potential inappropriate data. While this is well-established, we will investigate and demonstrate the promises of deep models’ acquired knowledge and capabilities through the provision of this very particular potentially inappropriate data. Importantly, we present the first approaches for learning ethics from data. Our findings suggest that if we build an AI system that learns an improved representation of data and that is able to better understand and produce it, in the process, it will also acquire more accurate societal knowledge, in this case, historical cultural associations to make human-like \"right\" and \"wrong\" choices. Furthermore, based on these findings, we consequently ask the arguably \"circular\" question of whether a machine can help us mitigate their associated concerns. Importantly, we demonstrate the importance of their ability to distinguish between \"right\" and \"wrong\" and how utilizing them can mitigate associated risks surrounding large-scale models themselves. However, we also highlight the role of human-machine interaction to explore and reinforce AI systems’ properties, including their flaws and merits, and present how human feedback on explanations can align deep learning based models with our precepts. We present these algorithms and corresponding findings, providing important insights for the goal of putting human values into AI systems, which, summarized, may not be insurmountable in the long run.\n

\n\n\n

\n \n\n \n \n \n \n \n \n Exploiting Cultural Biases via Homoglyphs in Text-to-Image Synthesis.\n \n \n \n \n\n\n \n Struppek, L.; Hintersdorf, D.; Friedrich, F.; Br, M.; Schramowski, P.; and Kersting, K.\n\n\n \n\n\n\n Journal of Artificial Intelligence Research, 78: 1017–1068. December 2023.\n \n\n\n\n
\n\n\n\n \n \n $\"Exploiting$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{struppekExploitingCulturalBiases2023,\n\ttitle = {Exploiting {Cultural} {Biases} via {Homoglyphs} in {Text}-to-{Image} {Synthesis}},\n\tvolume = {78},\n\tissn = {1076-9757},\n\turl = {http://www.jair.org/index.php/jair/article/view/15388},\n\tdoi = {10.1613/jair.1.15388},\n\tabstract = {Models for text-to-image synthesis, such as DALL-E 2 and Stable Diffusion, have recently drawn a lot of interest from academia and the general public. These models are capable of producing high-quality images that depict a variety of concepts and styles when conditioned on textual descriptions. However, these models adopt cultural characteristics associated with specific Unicode scripts from their vast amount of training data, which may not be immediately apparent. We show that by simply inserting single non-Latin characters in the textual description, common models reflect cultural biases in their generated images. We analyze this behavior both qualitatively and quantitatively and identify a model’s text encoder as the root cause of the phenomenon. Such behavior can be interpreted as a model feature, offering users a simple way to customize the image generation and reflect their own cultural background. Yet, malicious users or service providers may also try to intentionally bias the image generation. One goal might be to create racist stereotypes by replacing Latin characters with similarly-looking characters from non-Latin scripts, so-called homoglyphs. To mitigate such unnoticed script attacks, we propose a novel homoglyph unlearning method to fine-tune a text encoder, making it robust against homoglyph manipulations.},\n\turldate = {2025-05-13},\n\tjournal = {Journal of Artificial Intelligence Research},\n\tauthor = {Struppek, Lukas and Hintersdorf, Dom and Friedrich, Felix and Br, Manuel and Schramowski, Patrick and Kersting, Kristian},\n\tmonth = dec,\n\tyear = {2023},\n\tpages = {1017--1068},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n DSMC Evaluation Stages: Fostering Robust and Safe Behavior in Deep Reinforcement Learning – Extended Version.\n \n \n \n \n\n\n \n Gros, T. P.; Groß, J.; Höller, D.; Hoffmann, J.; Klauck, M.; Meerkamp, H.; Müller, N. J.; Schaller, L.; and Wolf, V.\n\n\n \n\n\n\n ACM Transactions on Modeling and Computer Simulation, 33(4): 1–28. October 2023.\n \n\n\n\n
\n\n\n\n \n \n $\"DSMC$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n \n \n 1 download\n \n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{grosDSMCEvaluationStages2023,\n\ttitle = {{DSMC} {Evaluation} {Stages}: {Fostering} {Robust} and {Safe} {Behavior} in {Deep} {Reinforcement} {Learning} – {Extended} {Version}},\n\tvolume = {33},\n\tissn = {1049-3301, 1558-1195},\n\tshorttitle = {{DSMC} {Evaluation} {Stages}},\n\turl = {https://dl.acm.org/doi/10.1145/3607198},\n\tdoi = {10.1145/3607198},\n\tabstract = {Neural networks (NN) are gaining importance in sequential decision-making. Deep reinforcement learning (DRL), in particular, is extremely successful in learning action policies in complex and dynamic environments. Despite this success, however, DRL technology is not without its failures, especially in safety-critical applications: (i) the training objective maximizes \n              average \n              rewards, which may disregard rare but critical situations and hence lack local robustness; (ii) optimization objectives targeting safety typically yield degenerated reward structures, which, for DRL to work, must be replaced with proxy objectives. Here, we introduce a methodology that can help to address both deficiencies. We incorporate \n              evaluation stages \n              (ES) into DRL, leveraging recent work on deep statistical model checking (DSMC), which verifies NN policies in Markov decision processes. Our ES apply DSMC at regular intervals to determine state space regions with weak performance. We adapt the subsequent DRL training priorities based on the outcome, (i) focusing DRL on critical situations and (ii) allowing to foster arbitrary objectives. \n             \n            We run case studies on two benchmarks. One of them is the Racetrack, an abstraction of autonomous driving that requires navigating a map without crashing into a wall. The other is MiniGrid, a widely used benchmark in the AI community. Our results show that DSMC-based ES can significantly improve both (i) and (ii).},\n\tlanguage = {en},\n\tnumber = {4},\n\turldate = {2025-05-13},\n\tjournal = {ACM Transactions on Modeling and Computer Simulation},\n\tauthor = {Gros, Timo P. and Groß, Joschka and Höller, Daniel and Hoffmann, Jörg and Klauck, Michaela and Meerkamp, Hendrik and Müller, Nicola J. and Schaller, Lukas and Wolf, Verena},\n\tmonth = oct,\n\tyear = {2023},\n\tpages = {1--28},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Analyzing neural network behavior through deep statistical model checking.\n \n \n \n \n\n\n \n Gros, T. P.; Hermanns, H.; Hoffmann, J.; Klauck, M.; and Steinmetz, M.\n\n\n \n\n\n\n International Journal on Software Tools for Technology Transfer, 25(3): 407–426. June 2023.\n \n\n\n\n
\n\n\n\n \n \n $\"Analyzing$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{grosAnalyzingNeuralNetwork2023,\n\ttitle = {Analyzing neural network behavior through deep statistical model checking},\n\tvolume = {25},\n\tissn = {1433-2779, 1433-2787},\n\turl = {https://link.springer.com/10.1007/s10009-022-00685-9},\n\tdoi = {10.1007/s10009-022-00685-9},\n\tabstract = {Abstract \n             \n              Neural networks (NN) are taking over ever more decisions thus far taken by humans, even though verifiable system-level guarantees are far out of reach. Neither is the verification technology available, nor is it even understood what a formal, meaningful, extensible, and scalable testbed might look like for such a technology. The present paper is an attempt to improve on both the above aspects. We present a family of formal models that contain basic features of automated decision-making contexts and which can be extended with further orthogonal features, ultimately encompassing the scope of autonomous driving. Due to the possibility to model random noise in the decision actuation, each model instance induces a Markov decision process (MDP) as verification object. The NN in this context has the duty to actuate (near-optimal) decisions. From the verification perspective, the externally learnt NN serves as a determinizer of the MDP, the result being a Markov chain which as such is amenable to statistical model checking. The combination of an MDP and an NN encoding the action policy is central to what we call “deep statistical model checking” (DSMC). While being a straightforward extension of statistical model checking, it enables to gain deep insight into questions like “how high is the NN-induced safety risk?”, “how good is the NN compared to the optimal policy?” (obtained by model checking the MDP), or “does further training improve the NN?”. We report on an implementation of DSMC inside the \n              Modest \n              Toolset \n              in combination with externally learnt NNs, demonstrating the potential of DSMC on various instances of the model family, and illustrating its scalability as a function of instance size as well as other factors like the degree of NN training.},\n\tlanguage = {en},\n\tnumber = {3},\n\turldate = {2025-05-13},\n\tjournal = {International Journal on Software Tools for Technology Transfer},\n\tauthor = {Gros, Timo P. and Hermanns, Holger and Hoffmann, Jörg and Klauck, Michaela and Steinmetz, Marcel},\n\tmonth = jun,\n\tyear = {2023},\n\tpages = {407--426},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Causing Intended Effects in Collaborative Decision-Making.\n \n \n \n \n\n\n \n Meyer-Vitali, A.; and Mulder, W.\n\n\n \n\n\n\n In Murukannaiah, P. K.; and Hirzle, T., editor(s), Proceedings of the Workshops at the Second International Conference on Hybrid Human-Artificial Intelligence, volume 3456, of CEUR Workshop Proceedings, pages 137–144, Munich, Germany, June 2023. CEUR\n \n\n\n\n
\n\n\n\n \n \n $\"Causing$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{meyer-vitaliCausingIntendedEffects2023,\n\taddress = {Munich, Germany},\n\tseries = {{CEUR} {Workshop} {Proceedings}},\n\ttitle = {Causing {Intended} {Effects} in {Collaborative} {Decision}-{Making}},\n\tvolume = {3456},\n\tcopyright = {All rights reserved},\n\tissn = {1613-0073},\n\turl = {https://ceur-ws.org/Vol-3456/#short4-1},\n\tlanguage = {en},\n\turldate = {2023-08-16},\n\tbooktitle = {Proceedings of the {Workshops} at the {Second} {International} {Conference} on {Hybrid} {Human}-{Artificial} {Intelligence}},\n\tpublisher = {CEUR},\n\tauthor = {Meyer-Vitali, André and Mulder, Wico},\n\teditor = {Murukannaiah, Pradeep K. and Hirzle, Teresa},\n\tmonth = jun,\n\tyear = {2023},\n\tpages = {137--144},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Knowledge Engineering for Hybrid Intelligence.\n \n \n \n \n\n\n \n Tiddi, I.; De Boer, V.; Schlobach, S.; and Meyer-Vitali, A.\n\n\n \n\n\n\n In Proceedings of the 12th Knowledge Capture Conference 2023, of K-CAP '23, pages 75–82, New York, NY, USA, December 2023. Association for Computing Machinery\n \n\n\n\n
\n\n\n\n \n \n $\"Knowledge$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n\n\n\n

@inproceedings{tiddiKnowledgeEngineeringHybrid2023,\n\taddress = {New York, NY, USA},\n\tseries = {K-{CAP} '23},\n\ttitle = {Knowledge {Engineering} for {Hybrid} {Intelligence}},\n\tcopyright = {All rights reserved},\n\tisbn = {979-8-4007-0141-2},\n\turl = {https://doi.org/10.1145/3587259.3627541},\n\tdoi = {10.1145/3587259.3627541},\n\tabstract = {Hybrid Intelligence (HI) is a rapidly growing field aiming at creating collaborative systems where humans and intelligent machines cooperate in mixed teams towards shared goals. A clear characterization of the tasks and knowledge exchanged by the agents in HI applications is still missing, hampering both standardization and reuse when designing new HI systems. Knowledge Engineering (KE) methods have been used to solve such issue through the formalization of tasks and roles in knowledge-intensive processes. We investigate whether KE methods can be applied to HI scenarios, and specifically whether common, reusable elements such as knowledge roles, tasks and subtasks can be identified in contexts where symbolic, subsymbolic and human-in-the-loop components are involved. We first adapt the well-known CommonKADS methodology to HI, and then use it to analyze several HI projects and identify common tasks. The results are (i) a high-level ontology of HI knowledge roles, (ii) a set of novel, HI-specific tasks and (iii) an open repository to store scenarios1 – allowing reuse, validation and design of existing and new HI applications.},\n\turldate = {2023-11-29},\n\tbooktitle = {Proceedings of the 12th {Knowledge} {Capture} {Conference} 2023},\n\tpublisher = {Association for Computing Machinery},\n\tauthor = {Tiddi, Ilaria and De Boer, Victor and Schlobach, Stefan and Meyer-Vitali, André},\n\tmonth = dec,\n\tyear = {2023},\n\tkeywords = {CommonKADS, Hybrid Intelligence, Knowledge Engineering},\n\tpages = {75--82},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Security of AI-Systems: Fundamentals - Security Considerations for Symbolic and Hybrid AI.\n \n \n \n \n\n\n \n Müller, C.; Vogt, R.; Nonnengart, A.; Klusch, M.; and Meyer-Vitali, A.\n\n\n \n\n\n\n Technical Report Bundesamt für Sicherheit in der Informationstechnik (BSI), June 2023.\n https://www.bsi.bund.de/DE/Service-Navi/Publikationen/Studien/Projekt_P464/Projekt_P464_node.html https://www.bsi.bund.de/DE/Service-Navi/Presse/Alle-Meldungen-News/Meldungen/Studien_Cyber-Sicherheit_KI-Systeme_230202.html\n\n\n\n
\n\n\n\n \n \n $\"Security$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@techreport{mullerSecurityAISystemsFundamentals2023,\n\ttitle = {Security of {AI}-{Systems}: {Fundamentals} - {Security} {Considerations} for {Symbolic} and {Hybrid} {AI}},\n\tcopyright = {All rights reserved},\n\tshorttitle = {Security of {AI}-{Systems}},\n\turl = {https://www.bsi.bund.de/SharedDocs/Downloads/EN/BSI/KI/Security-of-AI-systems_fundamentals_considerations_symbolic_hybrid.html},\n\tabstract = {Until recent years, cybersecurity was mostly concerned with an arms race of attackers and defence\nmechanisms on the level of admission control, protection of data transmission, cryptography, and so forth.\nOn the other side, functional safety of software-based systems dealt with faults in the system behaviour,\nwhich are caused by electro-mechanical errors in one of its sub-components or systematic errors (like bugs).\nWith AI-based systems becoming ever more wide-spread and complex, both paradigms need to be extended\nand, in a way, they are growing closer together. AI security and AI safety have a large overlap. Part of what\nAI safety tries to cope with are perturbations (or distribution shifts) that occur “naturally”, for example\nbecause the environment changes (day to night, summer to winter, Europe to Asia, simulation to reality,\netc.) or because the domain gradually evolves (demographic changes, generation changes, etc.). Seldom,\nevents can occur that have never been considered in training, causing an undesired emergent behaviour\n(misclassification, wrong decision, etc.) at inference time. Couldn’t we also say that the unexpected event\ncaused an “evasion”? AI security, aside from assuming an adversary, deals with similar problems. Data\npoisoning is the attempt to smuggle-in examples to the training set that decrease the accuracy of the system\n(or increase test error), thereby trying to be as efficient and subtle as possible. Evasion attacks alter the\ninference situation, either by manipulating the environment or otherwise making sure that the system\nreceives input that leads to misclassifications. In a sense, they are trying to create an event that was not\nexpected during training. It is possible that poisoning and evasion attacks are combined in a sense that the\npoisoning attack introduces a trigger for the later evasion. The proximity between the two problem domains\nexists on all levels: in highly automated driving, for example, it is plausible to describe a case in which the\ncar with ego-centric vision is tricked by the behaviour of another vehicle (agent) exhibiting a strange\nmanoeuvre. If we had reasons to assume that the agent’s “adversarial driving” was based on knowledge\nabout the inner working of the ego car, we would call it a security breach – otherwise a safety issue. It\nbecomes apparent that the distinction is somewhat arbitrary. Moreover, if we look at the body of literature\nin AI security, the game of finding new attacks, on the one side, and inventing new ways of defending them,\non the other, could also be framed under the umbrella of research on robustness.},\n\tinstitution = {Bundesamt für Sicherheit in der Informationstechnik (BSI)},\n\tauthor = {Müller, Christian and Vogt, Roland and Nonnengart, Andreas and Klusch, Matthias and Meyer-Vitali, André},\n\tmonth = jun,\n\tyear = {2023},\n\tnote = {https://www.bsi.bund.de/DE/Service-Navi/Publikationen/Studien/Projekt\\_P464/Projekt\\_P464\\_node.html\nhttps://www.bsi.bund.de/DE/Service-Navi/Presse/Alle-Meldungen-News/Meldungen/Studien\\_Cyber-Sicherheit\\_KI-Systeme\\_230202.html},\n}\n\n\n\n

\n\n\n

\n Until recent years, cybersecurity was mostly concerned with an arms race of attackers and defence mechanisms on the level of admission control, protection of data transmission, cryptography, and so forth. On the other side, functional safety of software-based systems dealt with faults in the system behaviour, which are caused by electro-mechanical errors in one of its sub-components or systematic errors (like bugs). With AI-based systems becoming ever more wide-spread and complex, both paradigms need to be extended and, in a way, they are growing closer together. AI security and AI safety have a large overlap. Part of what AI safety tries to cope with are perturbations (or distribution shifts) that occur “naturally”, for example because the environment changes (day to night, summer to winter, Europe to Asia, simulation to reality, etc.) or because the domain gradually evolves (demographic changes, generation changes, etc.). Seldom, events can occur that have never been considered in training, causing an undesired emergent behaviour (misclassification, wrong decision, etc.) at inference time. Couldn’t we also say that the unexpected event caused an “evasion”? AI security, aside from assuming an adversary, deals with similar problems. Data poisoning is the attempt to smuggle-in examples to the training set that decrease the accuracy of the system (or increase test error), thereby trying to be as efficient and subtle as possible. Evasion attacks alter the inference situation, either by manipulating the environment or otherwise making sure that the system receives input that leads to misclassifications. In a sense, they are trying to create an event that was not expected during training. It is possible that poisoning and evasion attacks are combined in a sense that the poisoning attack introduces a trigger for the later evasion. The proximity between the two problem domains exists on all levels: in highly automated driving, for example, it is plausible to describe a case in which the car with ego-centric vision is tricked by the behaviour of another vehicle (agent) exhibiting a strange manoeuvre. If we had reasons to assume that the agent’s “adversarial driving” was based on knowledge about the inner working of the ego car, we would call it a security breach – otherwise a safety issue. It becomes apparent that the distinction is somewhat arbitrary. Moreover, if we look at the body of literature in AI security, the game of finding new attacks, on the one side, and inventing new ways of defending them, on the other, could also be framed under the umbrella of research on robustness.\n

\n\n\n

\n \n\n \n \n \n \n \n A Maturity Model for Collaborative Agents in Human-AI Ecosystems.\n \n \n \n\n\n \n Mulder, W.; and Meyer-Vitali, A.\n\n\n \n\n\n\n In Camarinha-Matos, L. M.; Boucher, X.; and Ortiz, A., editor(s), Collaborative Networks in Digitalization and Society 5.0, of IFIP Advances in Information and Communication Technology, pages 328–335, Cham, 2023. Springer Nature Switzerland\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n\n\n\n

@inproceedings{mulderMaturityModelCollaborative2023,\n\taddress = {Cham},\n\tseries = {{IFIP} {Advances} in {Information} and {Communication} {Technology}},\n\ttitle = {A {Maturity} {Model} for {Collaborative} {Agents} in {Human}-{AI} {Ecosystems}},\n\tcopyright = {All rights reserved},\n\tisbn = {978-3-031-42622-3},\n\tdoi = {10.1007/978-3-031-42622-3_23},\n\tabstract = {AI entities lean on the aspects of their autonomy to carry out their tasks and perform intelligently. But when these entities collaborate in human-AI teams, their levels of autonomy and collaboration have to be balanced out. We present a maturity model for agents regarding this aspect of balancing. Whereas simple AI systems use pre-designed mechanisms, more advanced systems are able to learn this from experience. The maturity model is a two-dimensional matrix in which the degree of agency forms the horizontal axis, and the level of interaction the vertical axis. We validate the use of this maturity model with use-cases in the field of urban energy efficiency.},\n\tlanguage = {en},\n\tbooktitle = {Collaborative {Networks} in {Digitalization} and {Society} 5.0},\n\tpublisher = {Springer Nature Switzerland},\n\tauthor = {Mulder, Wico and Meyer-Vitali, André},\n\teditor = {Camarinha-Matos, Luis M. and Boucher, Xavier and Ortiz, Angel},\n\tyear = {2023},\n\tkeywords = {Agency, Collaborative networks, Human-AI teaming},\n\tpages = {328--335},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n

\n \n 2022\n \n \n (6)\n \n \n

\n \n \n

\n \n\n \n \n \n \n \n Comparing the Language of QAnon-related content on Parler, Gab, and Twitter.\n \n \n \n\n\n \n Sipka, A.; Hannak, A.; and Urman, A.\n\n\n \n\n\n\n In Proceedings of the 14th ACM Web Science Conference 2022, pages 411–421, 2022. \n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{sipkaComparingLanguageQAnonrelated2022,\n\ttitle = {Comparing the {Language} of {QAnon}-related content on {Parler}, {Gab}, and {Twitter}},\n\tbooktitle = {Proceedings of the 14th {ACM} {Web} {Science} {Conference} 2022},\n\tauthor = {Sipka, Andrea and Hannak, Aniko and Urman, Aleksandra},\n\tyear = {2022},\n\tpages = {411--421},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Conditional independence testing with heteroskedastic data and applications to causal discovery.\n \n \n \n\n\n \n Günther, W.; Ninad, U.; Wahl, J.; and Runge, J.\n\n\n \n\n\n\n Advances in Neural Information Processing Systems, 35: 16191–16202. 2022.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{guntherConditionalIndependenceTesting2022,\n\ttitle = {Conditional independence testing with heteroskedastic data and applications to causal discovery},\n\tvolume = {35},\n\tjournal = {Advances in Neural Information Processing Systems},\n\tauthor = {Günther, Wiebke and Ninad, Urmi and Wahl, Jonas and Runge, Jakob},\n\tyear = {2022},\n\tpages = {16191--16202},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Safe Latent Diffusion: Mitigating Inappropriate Degeneration in Diffusion Models.\n \n \n \n \n\n\n \n Schramowski, P.; Brack, M.; Deiseroth, B.; and Kersting, K.\n\n\n \n\n\n\n 2022.\n \n\n\n\n
\n\n\n\n \n \n $\"Safe$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n \n \n\n\n\n

@misc{schramowskiSafeLatentDiffusion2022,\n\ttitle = {Safe {Latent} {Diffusion}: {Mitigating} {Inappropriate} {Degeneration} in {Diffusion} {Models}},\n\tcopyright = {Creative Commons Attribution 4.0 International},\n\tshorttitle = {Safe {Latent} {Diffusion}},\n\turl = {https://arxiv.org/abs/2211.05105},\n\tdoi = {10.48550/ARXIV.2211.05105},\n\tabstract = {Text-conditioned image generation models have recently achieved astonishing results in image quality and text alignment and are consequently employed in a fast-growing number of applications. Since they are highly data-driven, relying on billion-sized datasets randomly scraped from the internet, they also suffer, as we demonstrate, from degenerated and biased human behavior. In turn, they may even reinforce such biases. To help combat these undesired side effects, we present safe latent diffusion (SLD). Specifically, to measure the inappropriate degeneration due to unfiltered and imbalanced training sets, we establish a novel image generation test bed-inappropriate image prompts (I2P)-containing dedicated, real-world image-to-text prompts covering concepts such as nudity and violence. As our exhaustive empirical evaluation demonstrates, the introduced SLD removes and suppresses inappropriate image parts during the diffusion process, with no additional training required and no adverse effect on overall image quality or text alignment.},\n\turldate = {2025-05-20},\n\tpublisher = {arXiv},\n\tauthor = {Schramowski, Patrick and Brack, Manuel and Deiseroth, Björn and Kersting, Kristian},\n\tyear = {2022},\n\tkeywords = {Artificial Intelligence (cs.AI), Computer Vision and Pattern Recognition (cs.CV), FOS: Computer and information sciences, Machine Learning (cs.LG)},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n From Responsibility to Reason-Giving Explainable Artificial Intelligence.\n \n \n \n\n\n \n Baum, K.; Mantel, S.; Schmidt, E.; and Speith, T.\n\n\n \n\n\n\n Philosophy & Technology, 35(1). 2022.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{baumResponsibilityReasonGivingExplainable2022,\n\ttitle = {From {Responsibility} to {Reason}-{Giving} {Explainable} {Artificial} {Intelligence}},\n\tvolume = {35},\n\tdoi = {10.1007/s13347-022-00510-w},\n\tnumber = {1},\n\tjournal = {Philosophy \\& Technology},\n\tauthor = {Baum, Kevin and Mantel, Susanne and Schmidt, Eva and Speith, Timo},\n\tyear = {2022},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Ethics for Nerds.\n \n \n \n\n\n \n Baum, K.; and Sterz, S.\n\n\n \n\n\n\n The International Review of Information Ethics, 31(1). 2022.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{baumEthicsNerds2022,\n\ttitle = {Ethics for {Nerds}},\n\tvolume = {31},\n\tdoi = {10.29173/irie484},\n\tnumber = {1},\n\tjournal = {The International Review of Information Ethics},\n\tauthor = {Baum, Kevin and Sterz, Sarah},\n\tyear = {2022},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Trustworthy Hybrid Team Decision-Support Systems.\n \n \n \n \n\n\n \n Meyer-Vitali, A.; and Mulder, W.\n\n\n \n\n\n\n . March 2022.\n Number: 7607\n\n\n\n
\n\n\n\n \n \n $\"Trustworthy$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{meyer-vitaliTrustworthyHybridTeam2022,\n\ttitle = {Trustworthy {Hybrid} {Team} {Decision}-{Support} {Systems}},\n\tcopyright = {All rights reserved},\n\tissn = {2516-2314},\n\turl = {https://easychair.org/publications/preprint/jRqf},\n\tlanguage = {en-US},\n\turldate = {2023-08-18},\n\tpublisher = {EasyChair},\n\tauthor = {Meyer-Vitali, André and Mulder, Wico},\n\tmonth = mar,\n\tyear = {2022},\n\tnote = {Number: 7607},\n}\n\n\n\n

\n\n\n\n

\n\n\n\n\n\n

\n\n

\n \n 2021\n \n \n (9)\n \n \n

\n \n \n

\n \n\n \n \n \n \n \n Explainability auditing for intelligent systems: a rationale for multi-disciplinary perspectives.\n \n \n \n\n\n \n Langer, M.; Baum, K.; Hartmann, K.; Hessel, S.; Speith, T.; and Wahl, J.\n\n\n \n\n\n\n In 2021 IEEE 29th international requirements engineering conference workshops (REW), pages 164–168, 2021. IEEE\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{langerExplainabilityAuditingIntelligent2021,\n\ttitle = {Explainability auditing for intelligent systems: a rationale for multi-disciplinary perspectives},\n\tbooktitle = {2021 {IEEE} 29th international requirements engineering conference workshops ({REW})},\n\tpublisher = {IEEE},\n\tauthor = {Langer, Markus and Baum, Kevin and Hartmann, Kathrin and Hessel, Stefan and Speith, Timo and Wahl, Jonas},\n\tyear = {2021},\n\tpages = {164--168},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Optimized locking for replication solutions.\n \n \n \n\n\n \n Wilkinson, J. P; and Sipka, A.\n\n\n \n\n\n\n October 2021.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@misc{wilkinsonOptimizedLockingReplication2021,\n\ttitle = {Optimized locking for replication solutions},\n\tauthor = {Wilkinson, John P and Sipka, Andrea},\n\tmonth = oct,\n\tyear = {2021},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Adaptive Rational Activations to Boost Deep Reinforcement Learning.\n \n \n \n \n\n\n \n Delfosse, Q.; Schramowski, P.; Mundt, M.; Molina, A.; and Kersting, K.\n\n\n \n\n\n\n In Proceedings of the International Conference on Representation Learning, 2021. \n \n\n\n\n
\n\n\n\n \n \n $\"Adaptive$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n\n\n\n

@inproceedings{delfosseAdaptiveRationalActivations2021,\n\ttitle = {Adaptive {Rational} {Activations} to {Boost} {Deep} {Reinforcement} {Learning}},\n\tcopyright = {Creative Commons Attribution 4.0 International},\n\turl = {https://arxiv.org/abs/2102.09407},\n\tdoi = {10.48550/ARXIV.2102.09407},\n\tabstract = {Latest insights from biology show that intelligence not only emerges from the connections between neurons but that individual neurons shoulder more computational responsibility than previously anticipated. This perspective should be critical in the context of constantly changing distinct reinforcement learning environments, yet current approaches still primarily employ static activation functions. In this work, we motivate why rationals are suitable for adaptable activation functions and why their inclusion into neural networks is crucial. Inspired by recurrence in residual networks, we derive a condition under which rational units are closed under residual connections and formulate a naturally regularised version: the recurrent-rational. We demonstrate that equipping popular algorithms with (recurrent-)rational activations leads to consistent improvements on Atari games, especially turning simple DQN into a solid approach, competitive to DDQN and Rainbow.},\n\turldate = {2025-05-21},\n\tbooktitle = {Proceedings of the {International} {Conference} on {Representation} {Learning}},\n\tauthor = {Delfosse, Quentin and Schramowski, Patrick and Mundt, Martin and Molina, Alejandro and Kersting, Kristian},\n\tyear = {2021},\n\tkeywords = {FOS: Computer and information sciences, Machine Learning (cs.LG)},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n What Do We Want from Explainable Artificial Intelligence (XAI)? – A Stakeholder Perspective on XAI and a Conceptual Model Guiding Interdisciplinary XAI Research.\n \n \n \n\n\n \n Langer, M.; Oster, D.; Speith, T.; Hermanns, H.; Kästner, L.; Schmidt, E.; Sesing, A.; and Baum, K.\n\n\n \n\n\n\n Artificial Intelligence, 296. 2021.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{langerWhatWeWant2021,\n\ttitle = {What {Do} {We} {Want} from {Explainable} {Artificial} {Intelligence} ({XAI})? – {A} {Stakeholder} {Perspective} on {XAI} and a {Conceptual} {Model} {Guiding} {Interdisciplinary} {XAI} {Research}},\n\tvolume = {296},\n\tdoi = {10.1016/j.artint.2021.103473},\n\tjournal = {Artificial Intelligence},\n\tauthor = {Langer, Markus and Oster, Daniel and Speith, Timo and Hermanns, Holger and Kästner, Lena and Schmidt, Eva and Sesing, Andreas and Baum, Kevin},\n\tyear = {2021},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Towards Perspicuity Requirements.\n \n \n \n\n\n \n Sterz, S.; Baum, K.; Lauber-Rönsberg, A.; and Hermanns, H.\n\n\n \n\n\n\n In Yue, T.; and Mirakhorli, M., editor(s), 29th IEEE International Requirements Engineering Conference Workshops (RE 2021 Workshops), Notre Dame, Indiana, USA, pages 159–163, September 2021. IEEE\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{sterzPerspicuityRequirements2021,\n\ttitle = {Towards {Perspicuity} {Requirements}},\n\tdoi = {10.1109/REW53955.2021.00029},\n\tbooktitle = {29th {IEEE} {International} {Requirements} {Engineering} {Conference} {Workshops} ({RE} 2021 {Workshops}), {Notre} {Dame}, {Indiana}, {USA}},\n\tpublisher = {IEEE},\n\tauthor = {Sterz, Sarah and Baum, Kevin and Lauber-Rönsberg, Anne and Hermanns, Holger},\n\teditor = {Yue, Tao and Mirakhorli, Mehdi},\n\tmonth = sep,\n\tyear = {2021},\n\tpages = {159--163},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Spare Me the Details: How the Type of Information About Automated Interviews Influences Applicant Reactions.\n \n \n \n\n\n \n Langer, M.; Baum, K.; König, C. J; Hähne, V.; Oster, D.; and Speith, T.\n\n\n \n\n\n\n International Journal of Selection and Assessment, 29(2): 154–169. 2021.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{langerSpareMeDetails2021,\n\ttitle = {Spare {Me} the {Details}: {How} the {Type} of {Information} {About} {Automated} {Interviews} {Influences} {Applicant} {Reactions}},\n\tvolume = {29},\n\tdoi = {10.1111/ijsa.12325},\n\tnumber = {2},\n\tjournal = {International Journal of Selection and Assessment},\n\tauthor = {Langer, Markus and Baum, Kevin and König, Cornelius J and Hähne, Viviane and Oster, Daniel and Speith, Timo},\n\tyear = {2021},\n\tpages = {154--169},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n What to Expect from Opening Up ‘Black Boxes’? Comparing Perceptions of Justice Between Human and Automated Agents.\n \n \n \n\n\n \n Schlicker, N.; Langer, M.; Ötting, S. K; Baum, K.; König, C. J; and Wallach, D.\n\n\n \n\n\n\n Computers in Human Behavior, 122. 2021.\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{schlickerWhatExpectOpening2021,\n\ttitle = {What to {Expect} from {Opening} {Up} ‘{Black} {Boxes}’? {Comparing} {Perceptions} of {Justice} {Between} {Human} and {Automated} {Agents}},\n\tvolume = {122},\n\tdoi = {10.1016/j.chb.2021.106837},\n\tjournal = {Computers in Human Behavior},\n\tauthor = {Schlicker, Nadine and Langer, Markus and Ötting, Sonja K and Baum, Kevin and König, Cornelius J and Wallach, Dieter},\n\tyear = {2021},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Modular design patterns for hybrid learning and reasoning systems.\n \n \n \n \n\n\n \n van Bekkum, M.; de Boer, M.; van Harmelen, F.; Meyer-Vitali, A.; and Teije, A. t.\n\n\n \n\n\n\n Applied Intelligence, 51(9): 6528–6546. September 2021.\n \n\n\n\n
\n\n\n\n \n \n $\"Modular$ Paper\n \n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@article{vanbekkumModularDesignPatterns2021,\n\ttitle = {Modular design patterns for hybrid learning and reasoning systems},\n\tvolume = {51},\n\tcopyright = {All rights reserved},\n\tissn = {1573-7497},\n\turl = {https://doi.org/10.1007/s10489-021-02394-3},\n\tdoi = {10.1007/s10489-021-02394-3},\n\tabstract = {The unification of statistical (data-driven) and symbolic (knowledge-driven) methods is widely recognized as one of the key challenges of modern AI. Recent years have seen a large number of publications on such hybrid neuro-symbolic AI systems. That rapidly growing literature is highly diverse, mostly empirical, and is lacking a unifying view of the large variety of these hybrid systems. In this paper, we analyze a large body of recent literature and we propose a set of modular design patterns for such hybrid, neuro-symbolic systems. We are able to describe the architecture of a very large number of hybrid systems by composing only a small set of elementary patterns as building blocks. The main contributions of this paper are: 1) a taxonomically organised vocabulary to describe both processes and data structures used in hybrid systems; 2) a set of 15+ design patterns for hybrid AI systems organized in a set of elementary patterns and a set of compositional patterns; 3) an application of these design patterns in two realistic use-cases for hybrid AI systems. Our patterns reveal similarities between systems that were not recognized until now. Finally, our design patterns extend and refine Kautz’s earlier attempt at categorizing neuro-symbolic architectures.},\n\tlanguage = {en},\n\tnumber = {9},\n\turldate = {2022-03-29},\n\tjournal = {Applied Intelligence},\n\tauthor = {van Bekkum, Michael and de Boer, Maaike and van Harmelen, Frank and Meyer-Vitali, André and Teije, Annette ten},\n\tmonth = sep,\n\tyear = {2021},\n\tpages = {6528--6546},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Modular Design Patterns for Hybrid Actors.\n \n \n \n \n\n\n \n Meyer-Vitali, A.; Mulder, W.; and de Boer, M. H. T.\n\n\n \n\n\n\n In Cooperative AI Workshop, volume 2021, of NeurIPS, December 2021. \n arXiv: 2109.09331\n\n\n\n
\n\n\n\n \n \n $\"Modular$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n \n \n abstract \n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n \n \n \n\n\n\n

@inproceedings{meyer-vitaliModularDesignPatterns2021,\n\tseries = {{NeurIPS}},\n\ttitle = {Modular {Design} {Patterns} for {Hybrid} {Actors}},\n\tvolume = {2021},\n\tcopyright = {Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International Licence (CC-BY-NC-ND)},\n\turl = {http://arxiv.org/abs/2109.09331},\n\tabstract = {Recently, a boxology (graphical language) with design patterns for hybrid AI was proposed, combining symbolic and sub-symbolic learning and reasoning. In this paper, we extend this boxology with actors and their interactions. The main contributions of this paper are: 1) an extension of the taxonomy to describe distributed hybrid AI systems with actors and interactions; and 2) showing examples using a few design patterns relevant in multi-agent systems and human-agent interaction.},\n\turldate = {2021-11-17},\n\tbooktitle = {Cooperative {AI} {Workshop}},\n\tauthor = {Meyer-Vitali, André and Mulder, Wico and de Boer, Maaike H. T.},\n\tmonth = dec,\n\tyear = {2021},\n\tnote = {arXiv: 2109.09331},\n\tkeywords = {Computer Science - Artificial Intelligence, Computer Science - Multiagent Systems, Computer Science - Software Engineering},\n}\n\n\n\n

\n\n\n\n\n\n

\n\n

\n \n 2020\n \n \n (4)\n \n \n

\n \n \n

\n \n\n \n \n \n \n \n Using AI to help healthcare professionals stay up-to-date with medical research.\n \n \n \n\n\n \n de Bie, K.; Kishore, N.; Rentsch, A.; Rosado, P.; and Sipka, A.\n\n\n \n\n\n\n In AI for Social Good Workshop, 2020. \n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{debieUsingAIHelp2020,\n\ttitle = {Using {AI} to help healthcare professionals stay up-to-date with medical research},\n\tbooktitle = {{AI} for {Social} {Good} {Workshop}},\n\tauthor = {de Bie, Kim and Kishore, Nishant and Rentsch, Anthony and Rosado, Pablo and Sipka, Andrea},\n\tyear = {2020},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Checking data integrity of data storage systems.\n \n \n \n\n\n \n Wilkinson, J.; and Sipka, A.\n\n\n \n\n\n\n January 2020.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@misc{wilkinsonCheckingDataIntegrity2020,\n\ttitle = {Checking data integrity of data storage systems},\n\tpublisher = {Google Patents},\n\tauthor = {Wilkinson, John and Sipka, Andrea},\n\tmonth = jan,\n\tyear = {2020},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Multiple point-in-time copies on a remote system.\n \n \n \n\n\n \n Sipka, A.; and Wilkinson, J. P\n\n\n \n\n\n\n October 2020.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@misc{sipkaMultiplePointintimeCopies2020,\n\ttitle = {Multiple point-in-time copies on a remote system},\n\tauthor = {Sipka, Andrea and Wilkinson, John P},\n\tmonth = oct,\n\tyear = {2020},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Point-in-time copy on a remote system.\n \n \n \n\n\n \n Sipka, A.; and Wilkinson, J. P\n\n\n \n\n\n\n March 2020.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@misc{sipkaPointintimeCopyRemote2020,\n\ttitle = {Point-in-time copy on a remote system},\n\tpublisher = {Google Patents},\n\tauthor = {Sipka, Andrea and Wilkinson, John P},\n\tmonth = mar,\n\tyear = {2020},\n}\n\n\n\n

\n\n\n\n

\n\n\n\n\n\n

\n\n

\n \n 2019\n \n \n (6)\n \n \n

\n \n \n

\n \n\n \n \n \n \n \n Self-describing volume ancestry for data synchronization.\n \n \n \n\n\n \n Sipka, A.; and Wilkinson, J. P\n\n\n \n\n\n\n December 2019.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@misc{sipkaSelfdescribingVolumeAncestry2019,\n\ttitle = {Self-describing volume ancestry for data synchronization},\n\tpublisher = {Google Patents},\n\tauthor = {Sipka, Andrea and Wilkinson, John P},\n\tmonth = dec,\n\tyear = {2019},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Collision detection at multi-node storage sites.\n \n \n \n\n\n \n Rostagni, F. C; Sipka, A.; and Wilkinson, J. P\n\n\n \n\n\n\n January 2019.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@misc{rostagniCollisionDetectionMultinode2019,\n\ttitle = {Collision detection at multi-node storage sites},\n\tpublisher = {Google Patents},\n\tauthor = {Rostagni, Florent C and Sipka, Andrea and Wilkinson, John P},\n\tmonth = jan,\n\tyear = {2019},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Handling node failure in multi-node data storage systems.\n \n \n \n\n\n \n Rostagni, F.; Sipka, A.; and Wilkinson, J.\n\n\n \n\n\n\n October 2019.\n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@misc{rostagniHandlingNodeFailure2019,\n\ttitle = {Handling node failure in multi-node data storage systems},\n\tpublisher = {Google Patents},\n\tauthor = {Rostagni, Florent and Sipka, Andrea and Wilkinson, John},\n\tmonth = oct,\n\tyear = {2019},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Anforderungen an die Erklärbarkeit maschinengestützter Entscheidungen.\n \n \n \n \n\n\n \n Sesing, A.; and Baum, K.\n\n\n \n\n\n\n In Taeger, J., editor(s), Die Macht der Daten und der Algorithmen – Regulierung von IT, IoT und KI. Tagungsband DSRI-Herbstakademie 2019, pages 435–449, 2019. \n \n\n\n\n
\n\n\n\n \n \n $\"Anforderungen$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{sesingAnforderungenErklarbarkeitMaschinengestutzter2019,\n\ttitle = {Anforderungen an die {Erklärbarkeit} maschinengestützter {Entscheidungen}},\n\turl = {http://olwir.de/?content=reihen/uebersicht&sort=tb&isbn=978-3-95599-061-9},\n\tbooktitle = {Die {Macht} der {Daten} und der {Algorithmen} – {Regulierung} von {IT}, {IoT} und {KI}. {Tagungsband} {DSRI}-{Herbstakademie} 2019},\n\tauthor = {Sesing, Andreas and Baum, Kevin},\n\teditor = {Taeger, Jürgen},\n\tyear = {2019},\n\tpages = {435--449},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Towards a framework combining machine ethics and machine explainability.\n \n \n \n\n\n \n Baum, K.; Hermanns, H.; and Speith, T.\n\n\n \n\n\n\n In Finkbeiner, B.; and Kleinberg, S., editor(s), Proceedings of the 3rd Workshop on Formal Reasoning about Causation, Responsibility, and Explanations in Science and Technology (CREST 2018), Thessaloniki, Greece, 21st April 2018, 2019. \n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{baumFrameworkCombiningMachine2019,\n\ttitle = {Towards a framework combining machine ethics and machine explainability},\n\tdoi = {10.4204/EPTCS.286.4},\n\tbooktitle = {Proceedings of the 3rd {Workshop} on {Formal} {Reasoning} about {Causation}, {Responsibility}, and {Explanations} in {Science} and {Technology} ({CREST} 2018), {Thessaloniki}, {Greece}, 21st {April} 2018},\n\tauthor = {Baum, Kevin and Hermanns, Holger and Speith, Timo},\n\teditor = {Finkbeiner, Bernd and Kleinberg, Samantha},\n\tyear = {2019},\n}\n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n Explainability as a Non-Functional Requirement.\n \n \n \n\n\n \n Köhl, M. A; Baum, K.; Langer, M.; Oster, D.; Speith, T.; and Bohlender, D.\n\n\n \n\n\n\n In 27th IEEE International Requirements Engineering Conference (RE 2019), Jeju Island, South Korea, pages 363–368, 2019. IEEE\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{kohlExplainabilityNonFunctionalRequirement2019,\n\ttitle = {Explainability as a {Non}-{Functional} {Requirement}},\n\tdoi = {10.1109/RE.2019.00046},\n\tbooktitle = {27th {IEEE} {International} {Requirements} {Engineering} {Conference} ({RE} 2019), {Jeju} {Island}, {South} {Korea}},\n\tpublisher = {IEEE},\n\tauthor = {Köhl, Maximilian A and Baum, Kevin and Langer, Markus and Oster, Daniel and Speith, Timo and Bohlender, Dimitri},\n\tyear = {2019},\n\tpages = {363--368},\n}\n\n\n\n

\n\n\n\n

\n\n\n\n\n\n

\n\n

\n \n 2018\n \n \n (1)\n \n \n

\n \n \n

\n \n\n \n \n \n \n \n \n From Machine Ethics To Machine Explainability and Back.\n \n \n \n \n\n\n \n Baum, K.; Hermanns, H.; and Speith, T.\n\n\n \n\n\n\n In International Symposium on Artificial Intelligence and Mathematics (ISAIM 2018), Fort Lauderdale, Florida, USA, January 2018. \n \n\n\n\n
\n\n\n\n \n \n $\"From$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{baumMachineEthicsMachine2018,\n\ttitle = {From {Machine} {Ethics} {To} {Machine} {Explainability} and {Back}},\n\turl = {https://isaim2018.cs.ou.edu/papers/ISAIM2018_Ethics_Baum_etal.pdf},\n\tbooktitle = {International {Symposium} on {Artificial} {Intelligence} and {Mathematics} ({ISAIM} 2018), {Fort} {Lauderdale}, {Florida}, {USA}},\n\tauthor = {Baum, Kevin and Hermanns, Holger and Speith, Timo},\n\tmonth = jan,\n\tyear = {2018},\n}\n\n\n\n

\n\n\n\n

\n\n\n\n\n\n

\n\n

\n \n 2017\n \n \n (1)\n \n \n

\n \n \n

\n \n\n \n \n \n \n \n Two Challenges for CI Trustworthiness and How to Address Them.\n \n \n \n\n\n \n Baum, K.; Köhl, M.; and Schmidt, E.\n\n\n \n\n\n\n In Pereira-Fariña, M.; and Reed, C., editor(s), Proceedings of the 1st Workshop on Explainable Computational Intelligence (XCI 2017), Dundee, United Kingdom, September 2017. Association for Computational Linguistics\n \n\n\n\n
\n\n\n\n \n\n \n \n doi\n \n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{baumTwoChallengesCI2017,\n\taddress = {Dundee, United Kingdom},\n\ttitle = {Two {Challenges} for {CI} {Trustworthiness} and {How} to {Address} {Them}},\n\tdoi = {10.18653/v1/W17-3701},\n\tbooktitle = {Proceedings of the 1st {Workshop} on {Explainable} {Computational} {Intelligence} ({XCI} 2017)},\n\tpublisher = {Association for Computational Linguistics},\n\tauthor = {Baum, Kevin and Köhl, Maximilian and Schmidt, Eva},\n\teditor = {Pereira-Fariña, M. and Reed, C.},\n\tmonth = sep,\n\tyear = {2017},\n}\n\n\n\n

\n\n\n\n

\n\n\n\n\n\n

\n\n

\n \n undefined\n \n \n (2)\n \n \n

\n \n \n

\n \n\n \n \n \n \n \n .\n \n \n \n\n\n \n \n\n\n \n\n\n\n \n \n\n\n\n
\n\n\n\n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n \n \n 9 downloads\n \n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

\n\n\n\n

\n\n\n

\n \n\n \n \n \n \n \n \n Structural Causal Bottleneck Models.\n \n \n \n \n\n\n \n Bing, S.; Wahl, J.; and Runge, J.\n\n\n \n\n\n\n In UAI 2025 Workshop on Causal Abstractions and Representations, . \n \n\n\n\n
\n\n\n\n \n \n $\"Structural$ Paper\n \n \n\n \n\n \n link\n \n \n\n bibtex\n \n\n \n\n \n\n \n \n \n \n \n \n \n\n \n \n \n\n\n\n

@inproceedings{bingStructuralCausalBottleneck,\n\ttitle = {Structural {Causal} {Bottleneck} {Models}},\n\turl = {https://openreview.net/forum?id=EFNvKMSITB},\n\turldate = {2026-02-26},\n\tbooktitle = {{UAI} 2025 {Workshop} on {Causal} {Abstractions} and {Representations}},\n\tauthor = {Bing, Simon and Wahl, Jonas and Runge, Jakob},\n}\n\n\n\n

\n\n\n\n

\n\n\n\n\n\n

\n\n\n\n\n

\n\n\n \n\n \n \n \n \n\n