Student-Oriented Teacher Knowledge Refinement for Knowledge Distillation

TL;DR

This paper proposes a student-oriented knowledge distillation method that refines the teacher’s knowledge and focuses transfer on mutual areas of interest, improving the effectiveness of knowledge transfer from large to small models.

Contribution

It introduces a novel student-oriented approach with a learnable feature augmentation and a mutual interest detection module, enhancing knowledge distillation effectiveness.

Findings

Improved student model performance across multiple benchmarks.

Enhanced focus on relevant knowledge areas during distillation.

Method is compatible with various existing distillation techniques.

Abstract

Knowledge distillation has become widely recognized for its ability to transfer knowledge from a large teacher network to a compact and more streamlined student network. Traditional knowledge distillation methods primarily follow a teacher-oriented paradigm that imposes the task of learning the teacher's complex knowledge onto the student network. However, significant disparities in model capacity and architectural design hinder the student's comprehension of the complex knowledge imparted by the teacher, resulting in sub-optimal performance. This paper introduces a novel perspective emphasizing student-oriented and refining the teacher's knowledge to better align with the student's needs, thereby improving knowledge transfer effectiveness. Specifically, we present the Student-Oriented Knowledge Distillation (SoKD), which incorporates a learnable feature augmentation strategy during training to refine the teacher's knowledge of the student dynamically. Furthermore, we deploy the Distinctive Area Detection Module (DAM) to identify areas of mutual interest between the teacher and student, concentrating knowledge transfer within these critical areas to avoid transferring irrelevant information. This customized module ensures a more focused and effective knowledge distillation process. Our approach, functioning as a plug-in, could be integrated with various knowledge distillation methods. Extensive experimental results demonstrate the efficacy and generalizability of our method.