Continual Learning with Dynamic Sparse Training: Exploring Algorithms for Effective Model Updates

Continual learning (CL) refers to the ability of an intelligent system to sequentially acquire and retain knowledge from a stream of data with as little computational overhead as possible. To this end; regularization, replay, architecture, and parameter isolation approaches were introduced to the literature. Parameter isolation using a sparse network which enables to allocate distinct parts of the neural network to different tasks and also allows to share of parameters between tasks if they are similar. Dynamic Sparse Training (DST) is a prominent way to find these sparse networks and isolate them for each task. This paper is the first empirical study investigating the effect of different DST components under the CL paradigm to fill a critical research gap and shed light on the optimal configuration of DST for CL if it exists. Therefore, we perform a comprehensive study in which we investigate various DST components to find the best topology per task on well-known CIFAR100 and miniImageNet benchmarks in a task-incremental CL setup since our primary focus is to evaluate the performance of various DST criteria, rather than the process of mask selection. We found that, at a low sparsity level, Erdos-Renyi Kernel (ERK) initialization utilizes the backbone more efficiently and allows to effectively learn increments of tasks. At a high sparsity level, however, uniform initialization demonstrates more reliable and robust performance. In terms of growth strategy; performance is dependent on the defined initialization strategy, and the extent of sparsity. Finally, adaptivity within DST components is a promising way for better continual learners.

Adaptive Regularization for Class-Incremental Learning

Class-Incremental Learning updates a deep classifier with new categories while maintaining the previously observed class accuracy. Regularizing the neural network weights is a common method to prevent forgetting previously learned classes while learning novel ones. However, existing regularizers use a constant magnitude throughout the learning sessions, which may not reflect the varying levels of difficulty of the tasks encountered during incremental learning. This study investigates the necessity of adaptive regularization in Class-Incremental Learning, which dynamically adjusts the regularization strength according to the complexity of the task at hand. We propose a Bayesian Optimization-based approach to automatically determine the optimal regularization magnitude for each learning task. Our experiments on two datasets via two regularizers demonstrate the importance of adaptive regularization for achieving accurate and less forgetful visual incremental learning.