BicKD: Bilateral Contrastive Knowledge Distillation

TL;DR

BicKD introduces a bilateral contrastive loss for knowledge distillation, enabling class-wise and sample-wise comparison, which improves transfer performance over traditional methods.

Contribution

This work proposes a novel bilateral contrastive loss for KD that emphasizes class-wise orthogonality and improves knowledge transfer effectiveness.

Findings

BicKD outperforms state-of-the-art KD methods across various benchmarks.

The bilateral contrastive loss enhances class separation and distribution structure.

Experiments demonstrate improved accuracy and transfer efficiency.

Abstract

Knowledge distillation (KD) is a machine learning framework that transfers knowledge from a teacher model to a student model. The vanilla KD proposed by Hinton et al. has been the dominant approach in logit-based distillation and demonstrates compelling performance. However, it only performs sample-wise probability alignment between teacher and student's predictions, lacking an mechanism for class-wise comparison. Besides, vanilla KD imposes no structural constraint on the probability space. In this work, we propose a simple yet effective methodology, bilateral contrastive knowledge distillation (BicKD). This approach introduces a novel bilateral contrastive loss, which intensifies the orthogonality among different class generalization spaces while preserving consistency within the same class. The bilateral formulation enables explicit comparison of both sample-wise and class-wise prediction patterns between teacher and student. By emphasizing probabilistic orthogonality, BicKD further regularizes the geometric structure of the predictive distribution. Extensive experiments show that our BicKD method enhances knowledge transfer, and consistently outperforms state-of-the-art knowledge distillation techniques across various model architectures and benchmarks.