Multi-Mode Online Knowledge Distillation for Self-Supervised Visual Representation Learning

TL;DR

This paper introduces MOKD, a novel multi-mode online knowledge distillation framework that enables two models to collaboratively improve self-supervised visual representations through self- and cross-distillation modes.

Contribution

The paper proposes a new MOKD method with self- and cross-distillation modes, including a cross-attention strategy, to enhance SSL performance without static teachers.

Findings

MOKD outperforms baseline models on various datasets.

Heterogeneous models benefit from MOKD in representation quality.

MOKD surpasses existing SSL-KD methods for both student and teacher models.

Abstract

Self-supervised learning (SSL) has made remarkable progress in visual representation learning. Some studies combine SSL with knowledge distillation (SSL-KD) to boost the representation learning performance of small models. In this study, we propose a Multi-mode Online Knowledge Distillation method (MOKD) to boost self-supervised visual representation learning. Different from existing SSL-KD methods that transfer knowledge from a static pre-trained teacher to a student, in MOKD, two different models learn collaboratively in a self-supervised manner. Specifically, MOKD consists of two distillation modes: self-distillation and cross-distillation modes. Among them, self-distillation performs self-supervised learning for each model independently, while cross-distillation realizes knowledge interaction between different models. In cross-distillation, a cross-attention feature search strategy is proposed to enhance the semantic feature alignment between different models. As a result, the two models can absorb knowledge from each other to boost their representation learning performance. Extensive experimental results on different backbones and datasets demonstrate that two heterogeneous models can benefit from MOKD and outperform their independently trained baseline. In addition, MOKD also outperforms existing SSL-KD methods for both the student and teacher models.