Understanding the Role of Training Regimes in Continual Learning

TL;DR

This paper investigates how training regimes like dropout, learning rate decay, and batch size influence catastrophic forgetting in neural networks, emphasizing the importance of local minima geometry for improved stability in continual learning.

Contribution

It introduces an analysis of how training hyperparameters affect local minima geometry and forgetting, offering practical techniques to enhance stability without altering the learning algorithm.

Findings

Dropout widens local minima, reducing forgetting.

Learning rate decay improves stability across tasks.

Optimal batch sizes help maintain knowledge over sequential tasks.

Abstract

Catastrophic forgetting affects the training of neural networks, limiting their ability to learn multiple tasks sequentially. From the perspective of the well established plasticity-stability dilemma, neural networks tend to be overly plastic, lacking the stability necessary to prevent the forgetting of previous knowledge, which means that as learning progresses, networks tend to forget previously seen tasks. This phenomenon coined in the continual learning literature, has attracted much attention lately, and several families of approaches have been proposed with different degrees of success. However, there has been limited prior work extensively analyzing the impact that different training regimes -- learning rate, batch size, regularization method-- can have on forgetting. In this work, we depart from the typical approach of altering the learning algorithm to improve stability. Instead, we hypothesize that the geometrical properties of the local minima found for each task play an important role in the overall degree of forgetting. In particular, we study the effect of dropout, learning rate decay, and batch size, on forming training regimes that widen the tasks' local minima and consequently, on helping it not to forget catastrophically. Our study provides practical insights to improve stability via simple yet effective techniques that outperform alternative baselines.