Utility-based Perturbed Gradient Descent: An Optimizer for Continual Learning
This addresses the challenge of fast adaptation in non-stationary environments for continual learning agents, though it appears incremental as it builds on existing gradient descent methods.
The paper tackles the problem of catastrophic forgetting and decaying plasticity in continual learning by proposing Utility-based Perturbed Gradient Descent (UPGD), an online algorithm that protects useful weights and perturbs less useful ones, resulting in reduced forgetting and maintained plasticity for modern representation learning methods.
Modern representation learning methods often struggle to adapt quickly under non-stationarity because they suffer from catastrophic forgetting and decaying plasticity. Such problems prevent learners from fast adaptation since they may forget useful features or have difficulty learning new ones. Hence, these methods are rendered ineffective for continual learning. This paper proposes Utility-based Perturbed Gradient Descent (UPGD), an online learning algorithm well-suited for continual learning agents. UPGD protects useful weights or features from forgetting and perturbs less useful ones based on their utilities. Our empirical results show that UPGD helps reduce forgetting and maintain plasticity, enabling modern representation learning methods to work effectively in continual learning.