EA-KD: Entropy-based Adaptive Knowledge Distillation

TL;DR

EA-KD introduces an entropy-based adaptive weighting scheme in knowledge distillation that dynamically emphasizes valuable samples, leading to improved performance across various tasks with minimal additional computational cost.

Contribution

It proposes a novel entropy-based method to adaptively reweight samples in knowledge distillation, enhancing effectiveness over traditional uniform approaches.

Findings

Consistently improves performance across image classification, object detection, and LLM distillation.

Achieves state-of-the-art results with negligible computational overhead.

Effectively prioritizes high-entropy samples for better knowledge transfer.

Abstract

Knowledge distillation (KD) enables a smaller "student" model to mimic a larger "teacher" model by transferring knowledge from the teacher's output or features. However, most KD methods treat all samples uniformly, overlooking the varying learning value of each sample and thereby limiting their effectiveness. In this paper, we propose Entropy-based Adaptive Knowledge Distillation (EA-KD), a simple yet effective plug-and-play KD method that prioritizes learning from valuable samples. EA-KD quantifies each sample's learning value by strategically combining the entropy of the teacher and student output, then dynamically reweights the distillation loss to place greater emphasis on high-entropy samples. Extensive experiments across diverse KD frameworks and tasks -- including image classification, object detection, and large language model (LLM) distillation -- demonstrate that EA-KD consistently enhances performance, achieving state-of-the-art results with negligible computational cost. Code is available at: https://github.com/cpsu00/EA-KD