On Explaining Knowledge Distillation: Measuring and Visualising the Knowledge Transfer Process

TL;DR

This paper introduces UniCAM, a gradient-based visualization method, and two metrics to interpret and quantify the knowledge transfer process in knowledge distillation, enhancing understanding of what the Student model learns from the Teacher.

Contribution

The paper presents UniCAM and two novel metrics, FSS and RS, to interpret and measure the effectiveness of knowledge transfer in KD, which was previously opaque.

Findings

UniCAM effectively visualizes the transfer of relevant features.

Distilled features focus on key input aspects, residual features target irrelevant areas.

Metrics FSS and RS quantify the relevance of transferred knowledge.

Abstract

Knowledge distillation (KD) remains challenging due to the opaque nature of the knowledge transfer process from a Teacher to a Student, making it difficult to address certain issues related to KD. To address this, we proposed UniCAM, a novel gradient-based visual explanation method, which effectively interprets the knowledge learned during KD. Our experimental results demonstrate that with the guidance of the Teacher's knowledge, the Student model becomes more efficient, learning more relevant features while discarding those that are not relevant. We refer to the features learned with the Teacher's guidance as distilled features and the features irrelevant to the task and ignored by the Student as residual features. Distilled features focus on key aspects of the input, such as textures and parts of objects. In contrast, residual features demonstrate more diffused attention, often targeting irrelevant areas, including the backgrounds of the target objects. In addition, we proposed two novel metrics: the feature similarity score (FSS) and the relevance score (RS), which quantify the relevance of the distilled knowledge. Experiments on the CIFAR10, ASIRRA, and Plant Disease datasets demonstrate that UniCAM and the two metrics offer valuable insights to explain the KD process.