Toward Student-Oriented Teacher Network Training For Knowledge Distillation

TL;DR

This paper proposes a novel teacher training method called SoTeacher, designed to optimize teacher performance specifically for knowledge distillation, leading to improved student accuracy across various datasets and algorithms.

Contribution

It introduces a theoretical framework linking teacher training to true label distribution approximation and proposes a practical training method incorporating Lipschitz and consistency regularization.

Findings

SoTeacher improves student accuracy consistently.

Theoretical analysis connects teacher training to label distribution approximation.

Empirical results validate the effectiveness of SoTeacher across datasets.

Abstract

How to conduct teacher training for knowledge distillation is still an open problem. It has been widely observed that a best-performing teacher does not necessarily yield the best-performing student, suggesting a fundamental discrepancy between the current teacher training practice and the ideal teacher training strategy. To fill this gap, we explore the feasibility of training a teacher that is oriented toward student performance with empirical risk minimization (ERM). Our analyses are inspired by the recent findings that the effectiveness of knowledge distillation hinges on the teacher's capability to approximate the true label distribution of training inputs. We theoretically establish that the ERM minimizer can approximate the true label distribution of training data as long as the feature extractor of the learner network is Lipschitz continuous and is robust to feature transformations. In light of our theory, we propose a teacher training method SoTeacher which incorporates Lipschitz regularization and consistency regularization into ERM. Experiments on benchmark datasets using various knowledge distillation algorithms and teacher-student pairs confirm that SoTeacher can improve student accuracy consistently.