English

ResKD: Residual-Guided Knowledge Distillation

Computer Vision and Pattern Recognition 2021-12-01 v4

Abstract

Knowledge distillation, aimed at transferring the knowledge from a heavy teacher network to a lightweight student network, has emerged as a promising technique for compressing neural networks. However, due to the capacity gap between the heavy teacher and the lightweight student, there still exists a significant performance gap between them. In this paper, we see knowledge distillation in a fresh light, using the knowledge gap, or the residual, between a teacher and a student as guidance to train a much more lightweight student, called a res-student. We combine the student and the res-student into a new student, where the res-student rectifies the errors of the former student. Such a residual-guided process can be repeated until the user strikes the balance between accuracy and cost. At inference time, we propose a sample-adaptive strategy to decide which res-students are not necessary for each sample, which can save computational cost. Experimental results show that we achieve competitive performance with 18.04%\%, 23.14%\%, 53.59%\%, and 56.86%\% of the teachers' computational costs on the CIFAR-10, CIFAR-100, Tiny-ImageNet, and ImageNet datasets. Finally, we do thorough theoretical and empirical analysis for our method.

Keywords

Cite

@article{arxiv.2006.04719,
  title  = {ResKD: Residual-Guided Knowledge Distillation},
  author = {Xuewei Li and Songyuan Li and Bourahla Omar and Fei Wu and Xi Li},
  journal= {arXiv preprint arXiv:2006.04719},
  year   = {2021}
}

Comments

The first two authors (Xuewei Li and Songyuan Li) contribute equally. Accepted to IEEE TRANSACTIONS ON IMAGE PROCESSING (TIP)

R2 v1 2026-06-23T16:09:08.441Z