AMD: Automatic Multi-step Distillation of Large-scale Vision Models

TL;DR

This paper introduces Automatic Multi-step Distillation (AMD), a novel method for effectively compressing large-scale vision models by progressively distilling through intermediate teacher-assistants, significantly improving performance over existing methods.

Contribution

The paper proposes a multi-step distillation framework with an automatic optimization process to identify optimal intermediate models for large-scale vision model compression.

Findings

AMD outperforms baseline distillation methods on CIFAR-10, CIFAR-100, and ImageNet.

The approach effectively handles large capacity gaps between teacher and student.

Extensive experiments validate the superiority of AMD in large-scale vision model compression.

Abstract

Transformer-based architectures have become the de-facto standard models for diverse vision tasks owing to their superior performance. As the size of the models continues to scale up, model distillation becomes extremely important in various real applications, particularly on devices limited by computational resources. However, prevailing knowledge distillation methods exhibit diminished efficacy when confronted with a large capacity gap between the teacher and the student, e.g, 10x compression rate. In this paper, we present a novel approach named Automatic Multi-step Distillation (AMD) for large-scale vision model compression. In particular, our distillation process unfolds across multiple steps. Initially, the teacher undergoes distillation to form an intermediate teacher-assistant model, which is subsequently distilled further to the student. An efficient and effective optimization framework is introduced to automatically identify the optimal teacher-assistant that leads to the maximal student performance. We conduct extensive experiments on multiple image classification datasets, including CIFAR-10, CIFAR-100, and ImageNet. The findings consistently reveal that our approach outperforms several established baselines, paving a path for future knowledge distillation methods on large-scale vision models.