Low-Rank Knowledge Decomposition for Medical Foundation Models

TL;DR

This paper introduces Low-Rank Knowledge Decomposition (LoRKD), a framework that enhances medical foundation models by decomposing them into specialized, lightweight experts, improving task-specific performance and transferability.

Contribution

The paper proposes a novel Low-Rank Knowledge Decomposition framework that explicitly separates knowledge into lightweight experts, improving specialization and resource efficiency in medical models.

Findings

Decomposed models outperform original models in performance.

The approach enhances transferability of medical foundation models.

Experimental results validate the effectiveness of LoRKD.

Abstract

The popularity of large-scale pre-training has promoted the development of medical foundation models. However, some studies have shown that although foundation models exhibit strong general feature extraction capabilities, their performance on specific tasks is still inferior to task-specific methods. In this paper, we explore a new perspective called ``Knowledge Decomposition'' to improve the performance on specific medical tasks, which deconstruct the foundation model into multiple lightweight expert models, each dedicated to a particular task, with the goal of improving specialization while concurrently mitigating resource expenditure. To accomplish the above objective, we design a novel framework named Low-Rank Knowledge Decomposition (LoRKD), which explicitly separates graidents by incorporating low-rank expert modules and the efficient knowledge separation convolution. Extensive experimental results demonstrate that the decomposed models perform well in terms of performance and transferability, even surpassing the original foundation models.