Abstract:Deep neural models have achieved remarkable performance on various supervised and unsupervised learning tasks, but it is a challenge to deploy these large-size networks on resource-limited devices. As a representative type of model compression and acceleration methods, knowledge distillation (KD) solves this problem by transferring knowledge from heavy teachers to lightweight students. However, most distillation methods focus on imitating the responses of teacher networks but ignore the information redundancy of student networks. In this article, we propose a novel distillation framework difference-based channel contrastive distillation (DCCD), which introduces channel contrastive knowledge and dynamic difference knowledge into student networks for redundancy reduction. At the feature level, we construct an efficient contrastive objective that broadens student networks' feature expression space and preserves richer information in the feature extraction stage. At the final output level, more detailed knowledge is extracted from teacher networks by making a difference between multiview augmented responses of the same instance. We enhance student networks to be more sensitive to minor dynamic changes. With the improvement of two aspects of DCCD, the student network gains contrastive and difference knowledge and reduces its overfitting and redundancy. Finally, we achieve surprising results that the student approaches and even outperforms the teacher in test accuracy on CIFAR-100. We reduce the top-1 error to 28.16% on ImageNet classification and 24.15% for cross-model transfer with ResNet-18. Empirical experiments and ablation studies on popular datasets show that our proposed method can achieve state-of-the-art accuracy compared with other distillation methods.

A Comprehensive Overhaul of Feature Distillation

DCCD: Reducing Neural Network Redundancy Via Distillation

Using Less but Important Information for Feature Distillation

Collaborative Distillation for Ultra-Resolution Universal Style Transfer

Show, Attend and Distill:Knowledge Distillation via Attention-based Feature Matching

DOT: A Distillation-Oriented Trainer

Task-Oriented Feature Distillation.

A Simple and Generic Framework for Feature Distillation via Channel-wise Transformation

Self-boosting for Feature Distillation

DFD: Distillng the Feature Disparity Differently for Detectors

Contrastive Representation Distillation

Improved Feature Distillation via Projector Ensemble

Pixel Distillation: Cost-flexible Distillation Across Image Sizes and Heterogeneous Networks

Customizing a Teacher for Feature Distillation

Pixel Distillation: A New Knowledge Distillation Scheme for Low-Resolution Image Recognition

Distilling Object Detectors via Decoupled Features

Controlling the Quality of Distillation in Response-Based Network Compression

DisWOT: Student Architecture Search for Distillation WithOut Training

Function-Consistent Feature Distillation

Normalized Feature Distillation for Semantic Segmentation

Improving Ensemble Distillation With Weight Averaging and Diversifying Perturbation