BioKD: Selective Physiology-to-Video Knowledge Distillation via Reliability Gate for Emotion Recognition. Hou, B., Li, R., Zhu, Y., Liu, H., Yu, L., & Wang, Y. August, 2026. Version Number: 1![BioKD: Selective Physiology-to-Video Knowledge Distillation via Reliability Gate for Emotion Recognition [link]](https://bibbase.org/img/filetypes/link.svg "link")
Paper doi abstract bibtex To address the limitations of video-based emotion recognition under ambiguous or socially masked behavioral cues, as well as the poor deployability of physiological signals, this paper proposes a reliability-aware physiology-to-video knowledge distillation framework, termed BioKD. The proposed framework leverages physiological signals as privileged information during training to guide a video-based student model in learning deep affective representations, while relying solely on non-intrusive video inputs at inference time. To cope with the high noise and instability of physiological teacher supervision caused by inter-subject variability, signal artifacts, and temporal inconsistency, BioKD incorporates a sample-wise reliability-aware gating mechanism together with a progressive distillation strategy. By adaptively regulating the strength of knowledge transfer, the framework suppresses negative transfer induced by unreliable physiological supervision and enables more stable cross-modal distillation. Experiments on DEAP and AMIGOS show that BioKD consistently outperforms representative baselines under both trial-wise and subject-wise evaluation protocols for valence and arousal recognition. For example, BioKD achieves 68.01\% on DEAP (trial-wise arousal) and 65.29\% under the more challenging subject-wise setting, demonstrating improved performance under a subject-independent evaluation setting. Further analyses show that BioKD effectively mitigates overconfident teacher errors and outperforms an entropy-only weighting strategy, confirming the importance of explicitly modeling supervision reliability. In addition, BioKD introduces no additional inference-time overhead relative to the same video student architecture and removes the need for physiological sensing and multimodal synchronization.
@misc{hou_biokd:_2026,
title = {{BioKD}: {Selective} {Physiology}-to-{Video} {Knowledge} {Distillation} via {Reliability} {Gate} for {Emotion} {Recognition}},
copyright = {Creative Commons Attribution 4.0 International},
shorttitle = {{BioKD}},
url = {https://arxiv.org/abs/2608.06023},
doi = {10.48550/ARXIV.2608.06023},
abstract = {To address the limitations of video-based emotion recognition under ambiguous or socially masked behavioral cues, as well as the poor deployability of physiological signals, this paper proposes a reliability-aware physiology-to-video knowledge distillation framework, termed BioKD. The proposed framework leverages physiological signals as privileged information during training to guide a video-based student model in learning deep affective representations, while relying solely on non-intrusive video inputs at inference time. To cope with the high noise and instability of physiological teacher supervision caused by inter-subject variability, signal artifacts, and temporal inconsistency, BioKD incorporates a sample-wise reliability-aware gating mechanism together with a progressive distillation strategy. By adaptively regulating the strength of knowledge transfer, the framework suppresses negative transfer induced by unreliable physiological supervision and enables more stable cross-modal distillation. Experiments on DEAP and AMIGOS show that BioKD consistently outperforms representative baselines under both trial-wise and subject-wise evaluation protocols for valence and arousal recognition. For example, BioKD achieves 68.01{\textbackslash}\% on DEAP (trial-wise arousal) and 65.29{\textbackslash}\% under the more challenging subject-wise setting, demonstrating improved performance under a subject-independent evaluation setting. Further analyses show that BioKD effectively mitigates overconfident teacher errors and outperforms an entropy-only weighting strategy, confirming the importance of explicitly modeling supervision reliability. In addition, BioKD introduces no additional inference-time overhead relative to the same video student architecture and removes the need for physiological sensing and multimodal synchronization.},
urldate = {2026-08-09},
publisher = {arXiv},
author = {Hou, Bojing and Li, Ruohao and Zhu, Yitong and Liu, Hongjun and Yu, Luwen and Wang, Yuyang},
month = aug,
year = {2026},
note = {Version Number: 1},
}
Downloads: 0
{"_id":"592W48tWymyiuZHsk","bibbaseid":"hou-li-zhu-liu-yu-wang-biokdselectivephysiologytovideoknowledgedistillationviareliabilitygateforemotionrecognition-2026","author_short":["Hou, B.","Li, R.","Zhu, Y.","Liu, H.","Yu, L.","Wang, Y."],"bibdata":{"bibtype":"misc","type":"misc","title":"BioKD: Selective Physiology-to-Video Knowledge Distillation via Reliability Gate for Emotion Recognition","copyright":"Creative Commons Attribution 4.0 International","shorttitle":"BioKD","url":"https://arxiv.org/abs/2608.06023","doi":"10.48550/ARXIV.2608.06023","abstract":"To address the limitations of video-based emotion recognition under ambiguous or socially masked behavioral cues, as well as the poor deployability of physiological signals, this paper proposes a reliability-aware physiology-to-video knowledge distillation framework, termed BioKD. The proposed framework leverages physiological signals as privileged information during training to guide a video-based student model in learning deep affective representations, while relying solely on non-intrusive video inputs at inference time. To cope with the high noise and instability of physiological teacher supervision caused by inter-subject variability, signal artifacts, and temporal inconsistency, BioKD incorporates a sample-wise reliability-aware gating mechanism together with a progressive distillation strategy. By adaptively regulating the strength of knowledge transfer, the framework suppresses negative transfer induced by unreliable physiological supervision and enables more stable cross-modal distillation. Experiments on DEAP and AMIGOS show that BioKD consistently outperforms representative baselines under both trial-wise and subject-wise evaluation protocols for valence and arousal recognition. For example, BioKD achieves 68.01\\% on DEAP (trial-wise arousal) and 65.29\\% under the more challenging subject-wise setting, demonstrating improved performance under a subject-independent evaluation setting. Further analyses show that BioKD effectively mitigates overconfident teacher errors and outperforms an entropy-only weighting strategy, confirming the importance of explicitly modeling supervision reliability. In addition, BioKD introduces no additional inference-time overhead relative to the same video student architecture and removes the need for physiological sensing and multimodal synchronization.","urldate":"2026-08-09","publisher":"arXiv","author":[{"propositions":[],"lastnames":["Hou"],"firstnames":["Bojing"],"suffixes":[]},{"propositions":[],"lastnames":["Li"],"firstnames":["Ruohao"],"suffixes":[]},{"propositions":[],"lastnames":["Zhu"],"firstnames":["Yitong"],"suffixes":[]},{"propositions":[],"lastnames":["Liu"],"firstnames":["Hongjun"],"suffixes":[]},{"propositions":[],"lastnames":["Yu"],"firstnames":["Luwen"],"suffixes":[]},{"propositions":[],"lastnames":["Wang"],"firstnames":["Yuyang"],"suffixes":[]}],"month":"August","year":"2026","note":"Version Number: 1","bibtex":"@misc{hou_biokd:_2026,\n\ttitle = {{BioKD}: {Selective} {Physiology}-to-{Video} {Knowledge} {Distillation} via {Reliability} {Gate} for {Emotion} {Recognition}},\n\tcopyright = {Creative Commons Attribution 4.0 International},\n\tshorttitle = {{BioKD}},\n\turl = {https://arxiv.org/abs/2608.06023},\n\tdoi = {10.48550/ARXIV.2608.06023},\n\tabstract = {To address the limitations of video-based emotion recognition under ambiguous or socially masked behavioral cues, as well as the poor deployability of physiological signals, this paper proposes a reliability-aware physiology-to-video knowledge distillation framework, termed BioKD. The proposed framework leverages physiological signals as privileged information during training to guide a video-based student model in learning deep affective representations, while relying solely on non-intrusive video inputs at inference time. To cope with the high noise and instability of physiological teacher supervision caused by inter-subject variability, signal artifacts, and temporal inconsistency, BioKD incorporates a sample-wise reliability-aware gating mechanism together with a progressive distillation strategy. By adaptively regulating the strength of knowledge transfer, the framework suppresses negative transfer induced by unreliable physiological supervision and enables more stable cross-modal distillation. Experiments on DEAP and AMIGOS show that BioKD consistently outperforms representative baselines under both trial-wise and subject-wise evaluation protocols for valence and arousal recognition. For example, BioKD achieves 68.01{\\textbackslash}\\% on DEAP (trial-wise arousal) and 65.29{\\textbackslash}\\% under the more challenging subject-wise setting, demonstrating improved performance under a subject-independent evaluation setting. Further analyses show that BioKD effectively mitigates overconfident teacher errors and outperforms an entropy-only weighting strategy, confirming the importance of explicitly modeling supervision reliability. In addition, BioKD introduces no additional inference-time overhead relative to the same video student architecture and removes the need for physiological sensing and multimodal synchronization.},\n\turldate = {2026-08-09},\n\tpublisher = {arXiv},\n\tauthor = {Hou, Bojing and Li, Ruohao and Zhu, Yitong and Liu, Hongjun and Yu, Luwen and Wang, Yuyang},\n\tmonth = aug,\n\tyear = {2026},\n\tnote = {Version Number: 1},\n}\n\n\n\n","author_short":["Hou, B.","Li, R.","Zhu, Y.","Liu, H.","Yu, L.","Wang, Y."],"key":"hou_biokd:_2026","id":"hou_biokd:_2026","bibbaseid":"hou-li-zhu-liu-yu-wang-biokdselectivephysiologytovideoknowledgedistillationviareliabilitygateforemotionrecognition-2026","role":"author","urls":{"Paper":"https://arxiv.org/abs/2608.06023"},"metadata":{"authorlinks":{}}},"bibtype":"misc","biburl":"https://bibbase.org/zotero-mypublications/yywangvr","dataSources":["jGKtB5DgeBbTCLaTG"],"keywords":[],"search_terms":["biokd","selective","physiology","video","knowledge","distillation","via","reliability","gate","emotion","recognition","hou","li","zhu","liu","yu","wang"],"title":"BioKD: Selective Physiology-to-Video Knowledge Distillation via Reliability Gate for Emotion Recognition","year":2026}