\n \n \n
\n
\n\n \n \n \n \n \n \n Asymptotic Analysis of Two-Layer Neural Networks after One Gradient Step under Gaussian Mixtures Data with Structure.\n \n \n \n \n\n\n \n Demir, S.; and Dogan, Z.\n\n\n \n\n\n\n In
The Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025, 2025. OpenReview.net\n
\n\n
\n\n
\n\n
\n\n \n \n
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
@inproceedings{DBLP:conf/iclr/DemirD25,\n author = {Samet Demir and\n Zafer Dogan},\n title = {Asymptotic Analysis of Two-Layer Neural Networks after One Gradient\n Step under Gaussian Mixtures Data with Structure},\n booktitle = {The Thirteenth International Conference on Learning Representations,\n {ICLR} 2025, Singapore, April 24-28, 2025},\n publisher = {OpenReview.net},\n year = {2025},\n url = {https://openreview.net/forum?id=tNn6Hskmti},\n timestamp = {Thu, 15 May 2025 01:00:00 +0200},\n biburl = {https://dblp.org/rec/conf/iclr/DemirD25.bib},\n bibsource = {dblp computer science bibliography, https://dblp.org}\n}\n\n\n
\n\n\n\n
\n\n\n
\n
\n\n \n \n \n \n \n \n Asymptotic Study of in-Context Learning with Random Transformers Through Equivalent Models.\n \n \n \n \n\n\n \n Demir, S.; and Dogan, Z.\n\n\n \n\n\n\n In
35th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2025, Istanbul, Turkey, August 31 - Sept. 3, 2025, pages 1–6, 2025. IEEE\n
\n\n
\n\n
\n\n
\n\n \n \n
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
\n
@inproceedings{DBLP:conf/mlsp/DemirD25,\n author = {Samet Demir and\n Zafer Dogan},\n title = {Asymptotic Study of in-Context Learning with Random Transformers Through\n Equivalent Models},\n booktitle = {35th {IEEE} International Workshop on Machine Learning for Signal\n Processing, {MLSP} 2025, Istanbul, Turkey, August 31 - Sept. 3, 2025},\n pages = {1--6},\n publisher = {{IEEE}},\n year = {2025},\n url = {https://doi.org/10.1109/MLSP62443.2025.11204336},\n doi = {10.1109/MLSP62443.2025.11204336},\n timestamp = {Thu, 06 Nov 2025 08:39:57 +0100},\n biburl = {https://dblp.org/rec/conf/mlsp/DemirD25.bib},\n bibsource = {dblp computer science bibliography, https://dblp.org}\n}\n\n\n
\n\n\n\n
\n\n\n
\n
\n\n \n \n \n \n \n \n Learning Rate Should Scale Inversely with High-Order Data Moments in High-Dimensional Online Independent Component Analysis.\n \n \n \n \n\n\n \n Gultekin, M. O.; Demir, S.; and Dogan, Z.\n\n\n \n\n\n\n In
35th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2025, Istanbul, Turkey, August 31 - Sept. 3, 2025, pages 1–6, 2025. IEEE\n
\n\n
\n\n
\n\n
\n\n \n \n
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
\n
@inproceedings{DBLP:conf/mlsp/GultekinDD25,\n author = {M. Oguzhan Gultekin and\n Samet Demir and\n Zafer Dogan},\n title = {Learning Rate Should Scale Inversely with High-Order Data Moments\n in High-Dimensional Online Independent Component Analysis},\n booktitle = {35th {IEEE} International Workshop on Machine Learning for Signal\n Processing, {MLSP} 2025, Istanbul, Turkey, August 31 - Sept. 3, 2025},\n pages = {1--6},\n publisher = {{IEEE}},\n year = {2025},\n url = {https://doi.org/10.1109/MLSP62443.2025.11204212},\n doi = {10.1109/MLSP62443.2025.11204212},\n timestamp = {Thu, 06 Nov 2025 00:00:00 +0100},\n biburl = {https://dblp.org/rec/conf/mlsp/GultekinDD25.bib},\n bibsource = {dblp computer science bibliography, https://dblp.org}\n}\n\n\n
\n\n\n\n
\n\n\n
\n
\n\n \n \n \n \n \n \n Benefits of Online Tilted Empirical Risk Minimization: A Case Study of Outlier Detection and Robust Regression.\n \n \n \n \n\n\n \n Yildirim, Y. E.; Demir, S.; and Dogan, Z.\n\n\n \n\n\n\n In
35th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2025, Istanbul, Turkey, August 31 - Sept. 3, 2025, pages 1–6, 2025. IEEE\n
\n\n
\n\n
\n\n
\n\n \n \n
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
\n
@inproceedings{DBLP:conf/mlsp/YildirimDD25,\n author = {Yigit E. Yildirim and\n Samet Demir and\n Zafer Dogan},\n title = {Benefits of Online Tilted Empirical Risk Minimization: {A} Case Study\n of Outlier Detection and Robust Regression},\n booktitle = {35th {IEEE} International Workshop on Machine Learning for Signal\n Processing, {MLSP} 2025, Istanbul, Turkey, August 31 - Sept. 3, 2025},\n pages = {1--6},\n publisher = {{IEEE}},\n year = {2025},\n url = {https://doi.org/10.1109/MLSP62443.2025.11204247},\n doi = {10.1109/MLSP62443.2025.11204247},\n timestamp = {Thu, 06 Nov 2025 00:00:00 +0100},\n biburl = {https://dblp.org/rec/conf/mlsp/YildirimDD25.bib},\n bibsource = {dblp computer science bibliography, https://dblp.org}\n}\n\n\n
\n\n\n\n
\n\n\n
\n
\n\n \n \n \n \n \n \n Asymptotic Analysis of Two-Layer Neural Networks after One Gradient Step under Gaussian Mixtures Data with Structure.\n \n \n \n \n\n\n \n Demir, S.; and Dogan, Z.\n\n\n \n\n\n\n
CoRR, abs/2503.00856. 2025.\n
\n\n
\n\n
\n\n
\n\n \n \n
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
\n
@article{DBLP:journals/corr/abs-2503-00856,\n author = {Samet Demir and\n Zafer Dogan},\n title = {Asymptotic Analysis of Two-Layer Neural Networks after One Gradient\n Step under Gaussian Mixtures Data with Structure},\n journal = {CoRR},\n volume = {abs/2503.00856},\n year = {2025},\n url = {https://doi.org/10.48550/arXiv.2503.00856},\n doi = {10.48550/ARXIV.2503.00856},\n eprinttype = {arXiv},\n eprint = {2503.00856},\n timestamp = {Mon, 21 Apr 2025 01:00:00 +0200},\n biburl = {https://dblp.org/rec/journals/corr/abs-2503-00856.bib},\n bibsource = {dblp computer science bibliography, https://dblp.org}\n}\n\n\n
\n\n\n\n
\n\n\n
\n
\n\n \n \n \n \n \n \n Learning Rate Should Scale Inversely with High-Order Data Moments in High-Dimensional Online Independent Component Analysis.\n \n \n \n \n\n\n \n Gultekin, M. O.; Demir, S.; and Dogan, Z.\n\n\n \n\n\n\n
CoRR, abs/2509.15127. 2025.\n
\n\n
\n\n
\n\n
\n\n \n \n
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
\n
@article{DBLP:journals/corr/abs-2509-15127,\n author = {M. Oguzhan Gultekin and\n Samet Demir and\n Zafer Dogan},\n title = {Learning Rate Should Scale Inversely with High-Order Data Moments\n in High-Dimensional Online Independent Component Analysis},\n journal = {CoRR},\n volume = {abs/2509.15127},\n year = {2025},\n url = {https://doi.org/10.48550/arXiv.2509.15127},\n doi = {10.48550/ARXIV.2509.15127},\n eprinttype = {arXiv},\n eprint = {2509.15127},\n timestamp = {Wed, 22 Oct 2025 01:00:00 +0200},\n biburl = {https://dblp.org/rec/journals/corr/abs-2509-15127.bib},\n bibsource = {dblp computer science bibliography, https://dblp.org}\n}\n\n\n
\n\n\n\n
\n\n\n
\n
\n\n \n \n \n \n \n \n Benefits of Online Tilted Empirical Risk Minimization: A Case Study of Outlier Detection and Robust Regression.\n \n \n \n \n\n\n \n Yildirim, Y. E.; Demir, S.; and Dogan, Z.\n\n\n \n\n\n\n
CoRR, abs/2509.15141. 2025.\n
\n\n
\n\n
\n\n
\n\n \n \n
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
\n
@article{DBLP:journals/corr/abs-2509-15141,\n author = {Yigit E. Yildirim and\n Samet Demir and\n Zafer Dogan},\n title = {Benefits of Online Tilted Empirical Risk Minimization: {A} Case Study\n of Outlier Detection and Robust Regression},\n journal = {CoRR},\n volume = {abs/2509.15141},\n year = {2025},\n url = {https://doi.org/10.48550/arXiv.2509.15141},\n doi = {10.48550/ARXIV.2509.15141},\n eprinttype = {arXiv},\n eprint = {2509.15141},\n timestamp = {Wed, 22 Oct 2025 01:00:00 +0200},\n biburl = {https://dblp.org/rec/journals/corr/abs-2509-15141.bib},\n bibsource = {dblp computer science bibliography, https://dblp.org}\n}\n\n\n
\n\n\n\n
\n\n\n
\n
\n\n \n \n \n \n \n \n Asymptotic Study of In-context Learning with Random Transformers through Equivalent Models.\n \n \n \n \n\n\n \n Demir, S.; and Dogan, Z.\n\n\n \n\n\n\n
CoRR, abs/2509.15152. 2025.\n
\n\n
\n\n
\n\n
\n\n \n \n
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
\n
@article{DBLP:journals/corr/abs-2509-15152,\n author = {Samet Demir and\n Zafer Dogan},\n title = {Asymptotic Study of In-context Learning with Random Transformers through\n Equivalent Models},\n journal = {CoRR},\n volume = {abs/2509.15152},\n year = {2025},\n url = {https://doi.org/10.48550/arXiv.2509.15152},\n doi = {10.48550/ARXIV.2509.15152},\n eprinttype = {arXiv},\n eprint = {2509.15152},\n timestamp = {Wed, 22 Oct 2025 01:00:00 +0200},\n biburl = {https://dblp.org/rec/journals/corr/abs-2509-15152.bib},\n bibsource = {dblp computer science bibliography, https://dblp.org}\n}\n\n\n
\n\n\n\n
\n\n\n
\n
\n\n \n \n \n \n \n \n How Data Mixing Shapes In-Context Learning: Asymptotic Equivalence for Transformers with MLPs.\n \n \n \n \n\n\n \n Demir, S.; and Dogan, Z.\n\n\n \n\n\n\n
CoRR, abs/2510.25753. 2025.\n
\n\n
\n\n
\n\n
\n\n \n \n
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
\n
@article{DBLP:journals/corr/abs-2510-25753,\n author = {Samet Demir and\n Zafer Dogan},\n title = {How Data Mixing Shapes In-Context Learning: Asymptotic Equivalence\n for Transformers with MLPs},\n journal = {CoRR},\n volume = {abs/2510.25753},\n year = {2025},\n url = {https://doi.org/10.48550/arXiv.2510.25753},\n doi = {10.48550/ARXIV.2510.25753},\n eprinttype = {arXiv},\n eprint = {2510.25753},\n timestamp = {Mon, 17 Nov 2025 00:00:00 +0100},\n biburl = {https://dblp.org/rec/journals/corr/abs-2510-25753.bib},\n bibsource = {dblp computer science bibliography, https://dblp.org}\n}\n\n\n
\n\n\n\n
\n\n\n
\n
\n\n \n \n \n \n \n \n Optimal Attention Temperature Enhances In-Context Learning under Distribution Shift.\n \n \n \n \n\n\n \n Demir, S.; and Dogan, Z.\n\n\n \n\n\n\n
CoRR, abs/2511.01292. 2025.\n
\n\n
\n\n
\n\n
\n\n \n \n
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
\n
@article{DBLP:journals/corr/abs-2511-01292,\n author = {Samet Demir and\n Zafer Dogan},\n title = {Optimal Attention Temperature Enhances In-Context Learning under Distribution\n Shift},\n journal = {CoRR},\n volume = {abs/2511.01292},\n year = {2025},\n url = {https://doi.org/10.48550/arXiv.2511.01292},\n doi = {10.48550/ARXIV.2511.01292},\n eprinttype = {arXiv},\n eprint = {2511.01292},\n timestamp = {Fri, 02 Jan 2026 00:00:00 +0100},\n biburl = {https://dblp.org/rec/journals/corr/abs-2511-01292.bib},\n bibsource = {dblp computer science bibliography, https://dblp.org}\n}\n\n\n
\n\n\n\n
\n\n\n
\n
\n\n \n \n \n \n \n \n Implicitly Normalized Online PCA: A Regularized Algorithm with Exact High-Dimensional Dynamics.\n \n \n \n \n\n\n \n Demir, S.; and Dogan, Z.\n\n\n \n\n\n\n
CoRR, abs/2512.01231. 2025.\n
\n\n
\n\n
\n\n
\n\n \n \n
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
\n
@article{DBLP:journals/corr/abs-2512-01231,\n author = {Samet Demir and\n Zafer Dogan},\n title = {Implicitly Normalized Online {PCA:} {A} Regularized Algorithm with\n Exact High-Dimensional Dynamics},\n journal = {CoRR},\n volume = {abs/2512.01231},\n year = {2025},\n url = {https://doi.org/10.48550/arXiv.2512.01231},\n doi = {10.48550/ARXIV.2512.01231},\n eprinttype = {arXiv},\n eprint = {2512.01231},\n timestamp = {Mon, 26 Jan 2026 00:00:00 +0100},\n biburl = {https://dblp.org/rec/journals/corr/abs-2512-01231.bib},\n bibsource = {dblp computer science bibliography, https://dblp.org}\n}\n\n\n
\n\n\n\n
\n\n\n\n\n\n