What is the role of memorization in Continual Learning?

2025-05-23

AI Summary

This study investigates the role of memory mechanisms in continual learning, uncovering an asymmetry between memorization and forgetting: highly memorized samples—though more prone to being forgotten in subsequent tasks—are critical for improving generalization, with their value increasing as the experience replay buffer expands. To address this, we propose the first computable proxy metric for sample-level memorization and design a dynamic buffer sampling strategy grounded in this metric. Through systematic incremental learning experiments, we demonstrate that prioritizing highly memorized samples significantly boosts overall performance under large-buffer regimes, whereas under tight memory constraints, retaining more representative (conventionally selected) samples yields better results. Our key contributions are threefold: (1) a theoretical characterization of the memorization–forgetting asymmetry; (2) a practical, differentiable framework for memorization estimation and memory-aware sampling; and (3) data-driven, buffer-size-dependent guidelines for optimal replay buffer management.

Technology Category

• Machine Learning: Life-Long and Continual Learning
• Search and Optimization: Learning to Search
• Multiagent Systems: Multiagent Learning

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: Social networks and social learning

Abstract

Memorization impacts the performance of deep learning algorithms. Prior works have studied memorization primarily in the context of generalization and privacy. This work studies the memorization effect on incremental learning scenarios. Forgetting prevention and memorization seem similar. However, one should discuss their differences. We designed extensive experiments to evaluate the impact of memorization on continual learning. We clarified that learning examples with high memorization scores are forgotten faster than regular samples. Our findings also indicated that memorization is necessary to achieve the highest performance. However, at low memory regimes, forgetting regular samples is more important. We showed that the importance of a high-memorization score sample rises with an increase in the buffer size. We introduced a memorization proxy and employed it in the buffer policy problem to showcase how memorization could be used during incremental training. We demonstrated that including samples with a higher proxy memorization score is beneficial when the buffer size is large.

Problem

• Examining memorization's role in continual learning performance
• Comparing memorization and forgetting prevention in incremental learning
• Evaluating memorization impact on sample retention and buffer policies

Innovation

• Studied memorization effect in continual learning
• Introduced memorization proxy for buffer policy
• High-memorization samples boost large buffer performance