BioKD: Selective Physiology-to-Video Knowledge Distillation via Reliability Gate for Emotion Recognition

2026-08-06

AI Summary

This work addresses the performance degradation of video-based emotion recognition when behavioral cues are ambiguous or deliberately masked, as well as the practical infeasibility of deploying physiological signals in real-world settings. To this end, the authors propose a reliability-aware knowledge distillation framework that leverages physiological signals as privileged information during training. By integrating a sample-level reliability gating mechanism with a progressive distillation strategy, the method effectively mitigates negative transfer caused by unreliable supervision and enhances representation learning through cross-modal alignment. Notably, at inference time, the model operates solely on video inputs without any additional computational overhead. Evaluated on the DEAP and AMIGOS datasets, the proposed approach significantly outperforms existing baselines, achieving an accuracy of 65.29% under the subject-wise evaluation protocol.

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.

Problem

• emotion recognition
• video-based
• physiological signals
• knowledge distillation
• reliability

Innovation

• knowledge distillation
• physiological signals
• reliability gating
• emotion recognition
• cross-modal learning