Toward Understanding Catastrophic Forgetting in Continual Learning

TL;DR

This paper investigates how properties of task sequences, like complexity and heterogeneity, influence catastrophic forgetting in continual learning, revealing strong correlations with complexity and surprising results with heterogeneity.

Contribution

The paper introduces a new procedure to analyze task sequence properties and their impact on continual learning error rates, focusing on total complexity and sequential heterogeneity.

Findings

Error rates are strongly correlated with total complexity for some algorithms.

Sequential heterogeneity has no or negative correlation with error rates in some cases.

The study provides insights for improving continual learning benchmarks and methods.

Abstract

We study the relationship between catastrophic forgetting and properties of task sequences. In particular, given a sequence of tasks, we would like to understand which properties of this sequence influence the error rates of continual learning algorithms trained on the sequence. To this end, we propose a new procedure that makes use of recent developments in task space modeling as well as correlation analysis to specify and analyze the properties we are interested in. As an application, we apply our procedure to study two properties of a task sequence: (1) total complexity and (2) sequential heterogeneity. We show that error rates are strongly and positively correlated to a task sequence's total complexity for some state-of-the-art algorithms. We also show that, surprisingly, the error rates have no or even negative correlations in some cases to sequential heterogeneity. Our findings suggest directions for improving continual learning benchmarks and methods.