Distilling Lightweight Domain Experts from Large ML Models by Identifying Relevant Subspaces

TL;DR

This paper introduces 'SubDistill', a novel knowledge distillation method that selectively transfers relevant subspace components from large models to smaller ones, improving efficiency and interpretability for specific subtasks.

Contribution

The paper proposes 'SubDistill', a new distillation algorithm that focuses on relevant subspaces, enhancing performance and interpretability in targeted tasks.

Findings

SubDistill outperforms existing distillation methods on CIFAR-100 and ImageNet.

Distilled models better match teacher decision structures.

Method improves efficiency by focusing on relevant components.

Abstract

Knowledge distillation involves transferring the predictive capabilities of large, high-performing AI models (teachers) to smaller models (students) that can operate in environments with limited computing power. In this paper, we address the scenario in which only a few classes and their associated intermediate concepts are relevant to distill. This scenario is common in practice, yet few existing distillation methods explicitly focus on the relevant subtask. To address this gap, we introduce 'SubDistill', a new distillation algorithm with improved numerical properties that only distills the relevant components of the teacher model at each layer. Experiments on CIFAR-100 and ImageNet with Convolutional and Transformer models demonstrate that SubDistill outperforms existing layer-wise distillation techniques on a representative set of subtasks. Our benchmark evaluations are complemented by Explainable AI analyses showing that our distilled student models more closely match the decision structure of the original teacher model.