Modular-Relatedness for Continual Learning

TL;DR

This paper introduces a modular-relatedness technique for continual learning that enhances task retention and reduces forgetting by automatically extracting neural network modules and estimating task relatedness, applicable across various CL methods.

Contribution

It presents a novel modular-relatedness approach that improves continual learning performance, especially under limited memory conditions, by enhancing task relatedness estimation.

Findings

Significant reduction in catastrophic forgetting.

Improved robustness of EWC and GEM methods.

Effective with limited episodic memory.

Abstract

In this paper, we propose a continual learning (CL) technique that is beneficial to sequential task learners by improving their retained accuracy and reducing catastrophic forgetting. The principal target of our approach is the automatic extraction of modular parts of the neural network and then estimating the relatedness between the tasks given these modular components. This technique is applicable to different families of CL methods such as regularization-based (e.g., the Elastic Weight Consolidation) or the rehearsal-based (e.g., the Gradient Episodic Memory) approaches where episodic memory is needed. Empirical results demonstrate remarkable performance gain (in terms of robustness to forgetting) for methods such as EWC and GEM based on our technique, especially when the memory budget is very limited.