Towards a Unified View of Affinity-Based Knowledge Distillation

TL;DR

This paper presents a unified framework for affinity-based knowledge distillation in neural networks, analyzing various combinations of modules to improve understanding and performance in image classification tasks.

Contribution

It introduces a modular framework for relation-based knowledge distillation, enabling systematic evaluation and revealing effective design choices.

Findings

Relation-based distillation can achieve performance comparable to state-of-the-art methods.

A modular framework clarifies the roles of affinity, normalization, and loss in distillation.

Certain module combinations lead to more effective knowledge transfer.

Abstract

Knowledge transfer between artificial neural networks has become an important topic in deep learning. Among the open questions are what kind of knowledge needs to be preserved for the transfer, and how it can be effectively achieved. Several recent work have shown good performance of distillation methods using relation-based knowledge. These algorithms are extremely attractive in that they are based on simple inter-sample similarities. Nevertheless, a proper metric of affinity and use of it in this context is far from well understood. In this paper, by explicitly modularising knowledge distillation into a framework of three components, i.e. affinity, normalisation, and loss, we give a unified treatment of these algorithms as well as study a number of unexplored combinations of the modules. With this framework we perform extensive evaluations of numerous distillation objectives for image classification, and obtain a few useful insights for effective design choices while demonstrating how relation-based knowledge distillation could achieve comparable performance to the state of the art in spite of the simplicity.