Multi-Teacher Knowledge Distillation with Reinforcement Learning for Visual Recognition
Chuanguang Yang, XinQiang Yu, Han Yang, Zhulin An, Chengqing Yu, Libo Huang, Yongjun Xu
Abstract
Multi-teacher Knowledge Distillation (KD) transfers diverse knowledge from a teacher pool to a student network. The core problem of multi-teacher KD is how to balance distillation strengths among various teachers. Most existing methods often develop weighting strategies from an individual perspective of teacher performance or teacher-student gaps, lacking comprehensive information for guidance. This paper proposes Multi-Teacher Knowledge Distillation with Reinforcement Learning (MTKD-RL) to optimize multi-teacher weights. In this framework, we construct both teacher performance and teacher-student gaps as state information to an agent. The agent outputs the teacher weight and can be updated by the return reward from the student. MTKD-RL reinforces the interaction between the student and teacher using an agent in an RL-based decision mechanism, achieving better matching capability with more meaningful weights. Experimental results on visual recognition tasks, including image classification, object detection, and semantic segmentation tasks, demonstrate that MTKD-RL achieves state-of-the-art performance compared to the existing multi-teacher KD works.
BibTeX
@article{Yang_Yu_Yang_An_Yu_Huang_Xu_2025, title={Multi-Teacher Knowledge Distillation with Reinforcement Learning for Visual Recognition}, volume={39}, url={https://ojs.aaai.org/index.php/AAAI/article/view/32990}, DOI={10.1609/aaai.v39i9.32990}, abstractNote={Multi-teacher Knowledge Distillation (KD) transfers diverse knowledge from a teacher pool to a student network. The core problem of multi-teacher KD is how to balance distillation strengths among various teachers. Most existing methods often develop weighting strategies from an individual perspective of teacher performance or teacher-student gaps, lacking comprehensive information for guidance. This paper proposes Multi-Teacher Knowledge Distillation with Reinforcement Learning (MTKD-RL) to optimize multi-teacher weights. In this framework, we construct both teacher performance and teacher-student gaps as state information to an agent. The agent outputs the teacher weight and can be updated by the return reward from the student. MTKD-RL reinforces the interaction between the student and teacher using an agent in an RL-based decision mechanism, achieving better matching capability with more meaningful weights. Experimental results on visual recognition tasks, including image classification, object detection, and semantic segmentation tasks, demonstrate that MTKD-RL achieves state-of-the-art performance compared to the existing multi-teacher KD works.}, number={9}, journal={Proceedings of the AAAI Conference on Artificial Intelligence}, author={Yang, Chuanguang and Yu, XinQiang and Yang, Han and An, Zhulin and Yu, Chengqing and Huang, Libo and Xu, Yongjun}, year={2025}, month={Apr.}, pages={9148-9156} }