Understanding the Role of Rehearsal Scale in Continual Learning under Varying Model Capacities

2026-02-24

AI Summary

This study investigates the impact of replay buffer size on model adaptability, memory retention, and generalization in continual learning, with particular emphasis on its non-monotonic effects across varying model capacities. Framing replay-based continual learning as a multidimensional efficacy-driven iterative optimization problem, the work unifies the learning dynamics underlying diverse performance metrics. Through closed-form theoretical analysis, numerical simulations, and experiments with deep neural networks on multiple real-world datasets, the study reveals counterintuitive phenomena—such as replay potentially impairing adaptability and larger replay buffers not necessarily improving memory retention—and derives a lower bound on memory error. This bound is shown to be jointly determined by replay size, task similarity, and noise level, thereby delineating the effective regime and applicability conditions of replay mechanisms.

Technology Category

• Machine Learning: Life-Long and Continual Learning
• Natural Language Processing: Learning & Optimization for NLP
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Practical large-scale studies of user experience
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Rehearsal is one of the key techniques for mitigating catastrophic forgetting and has been widely adopted in continual learning algorithms due to its simplicity and practicality. However, the theoretical understanding of how rehearsal scale influences learning dynamics remains limited. To address this gap, we formulate rehearsal-based continual learning as a multidimensional effectiveness-driven iterative optimization problem, providing a unified characterization across diverse performance metrics. Within this framework, we derive a closed-form analysis of adaptability, memorability, and generalization from the perspective of rehearsal scale. Our results uncover several intriguing and counterintuitive findings. First, rehearsal can impair model's adaptability, in sharp contrast to its traditionally recognized benefits. Second, increasing the rehearsal scale does not necessarily improve memory retention. When tasks are similar and noise levels are low, the memory error exhibits a diminishing lower bound. Finally, we validate these insights through numerical simulations and extended analyses on deep neural networks across multiple real-world datasets, revealing statistical patterns of rehearsal mechanisms in continual learning.

Problem

• rehearsal scale
• continual learning
• catastrophic forgetting
• model capacity
• learning dynamics

Innovation

• rehearsal scale
• continual learning
• catastrophic forgetting
• closed-form analysis
• model capacity