Knowledge distillation: A good teacher is patient and consistent

TL;DR

This paper investigates how to effectively use knowledge distillation to make large-scale computer vision models more practical by identifying key design choices that impact performance.

Contribution

It explicitly identifies critical design choices in knowledge distillation and provides a comprehensive empirical study demonstrating their importance.

Findings

Knowledge distillation can significantly reduce model size without performance loss.

Implicit design choices greatly influence distillation effectiveness.

Achieved state-of-the-art 82.8% top-1 accuracy with ResNet-50 on ImageNet.

Abstract

There is a growing discrepancy in computer vision between large-scale models that achieve state-of-the-art performance and models that are affordable in practical applications. In this paper we address this issue and significantly bridge the gap between these two types of models. Throughout our empirical investigation we do not aim to necessarily propose a new method, but strive to identify a robust and effective recipe for making state-of-the-art large scale models affordable in practice. We demonstrate that, when performed correctly, knowledge distillation can be a powerful tool for reducing the size of large models without compromising their performance. In particular, we uncover that there are certain implicit design choices, which may drastically affect the effectiveness of distillation. Our key contribution is the explicit identification of these design choices, which were not previously articulated in the literature. We back up our findings by a comprehensive empirical study, demonstrate compelling results on a wide range of vision datasets and, in particular, obtain a state-of-the-art ResNet-50 model for ImageNet, which achieves 82.8% top-1 accuracy.