Local vs Global continual learning

TL;DR

This paper compares local and global approximation strategies in continual learning, analyzing their mechanisms, classifying existing algorithms, and exploring optimal objectives for local polynomial approximations.

Contribution

It introduces a new perspective on continual learning through multi-task loss approximation and classifies algorithms based on approximation type.

Findings

Local and global strategies have distinct advantages and limitations.

Classified existing algorithms according to their approximation approach.

Identified optimal objectives for local polynomial approximations.

Abstract

Continual learning is the problem of integrating new information in a model while retaining the knowledge acquired in the past. Despite the tangible improvements achieved in recent years, the problem of continual learning is still an open one. A better understanding of the mechanisms behind the successes and failures of existing continual learning algorithms can unlock the development of new successful strategies. In this work, we view continual learning from the perspective of the multi-task loss approximation, and we compare two alternative strategies, namely local and global approximations. We classify existing continual learning algorithms based on the approximation used, and we assess the practical effects of this distinction in common continual learning settings.Additionally, we study optimal continual learning objectives in the case of local polynomial approximations and we provide examples of existing algorithms implementing the optimal objectives