DisWOT: Student Architecture Search for Distillation WithOut Training

TL;DR

DisWOT introduces a training-free method to search for optimal student architectures for knowledge distillation by measuring feature similarity, significantly reducing training time while improving performance.

Contribution

The paper proposes a novel training-free architecture search framework for distillation that correlates feature similarity with distillation performance, eliminating the need for training during search.

Findings

Achieves state-of-the-art results on CIFAR, ImageNet, NAS-Bench-201.

Significantly accelerates architecture search by at least 180×.

Effectively extends similarity metrics as new distillers and zero-proxies.

Abstract

Knowledge distillation (KD) is an effective training strategy to improve the lightweight student models under the guidance of cumbersome teachers. However, the large architecture difference across the teacher-student pairs limits the distillation gains. In contrast to previous adaptive distillation methods to reduce the teacher-student gap, we explore a novel training-free framework to search for the best student architectures for a given teacher. Our work first empirically show that the optimal model under vanilla training cannot be the winner in distillation. Secondly, we find that the similarity of feature semantics and sample relations between random-initialized teacher-student networks have good correlations with final distillation performances. Thus, we efficiently measure similarity matrixs conditioned on the semantic activation maps to select the optimal student via an evolutionary algorithm without any training. In this way, our student architecture search for Distillation WithOut Training (DisWOT) significantly improves the performance of the model in the distillation stage with at least 180$\times$ training acceleration. Additionally, we extend similarity metrics in DisWOT as new distillers and KD-based zero-proxies. Our experiments on CIFAR, ImageNet and NAS-Bench-201 demonstrate that our technique achieves state-of-the-art results on different search spaces. Our project and code are available at https://lilujunai.github.io/DisWOT-CVPR2023/.