Dynamic Dual Buffer with Divide-and-Conquer Strategy for Online Continual Learning

2025/05/23 by Dai, Congren, Zhou, Huichi, Huang, Jiahao +5
#FOS: Computer and information sciences #Machine Learning (cs.LG)

paper · doi:10.48550/arxiv.2505.18101

Abstract

Online Continual Learning (OCL) involves sequentially arriving data and is particularly challenged by catastrophic forgetting, which significantly impairs model performance. To address this issue, we introduce a novel framework, Online Dynamic Expandable Dual Memory (ODEDM), that integrates a short-term memory for fast memory and a long-term memory structured into sub-buffers anchored by cluster prototypes, enabling the storage of diverse and category-specific samples to mitigate forgetting. We propose a novel K-means-based strategy for prototype identification and an optimal transport-based mechanism to retain critical samples, prioritising those exhibiting high similarity to their corresponding prototypes. This design preserves semantically rich information. Additionally, we propose a Divide-and-Conquer (DAC) optimisation strategy that decomposes memory updates into subproblems, thereby reducing computational overhead. ODEDM functions as a plug-and-play module that can be seamlessly integrated with existing rehearsal-based approaches. Experimental results under both standard and imbalanced OCL settings show that ODEDM consistently achieves state-of-the-art performance across multiple datasets, delivering substantial improvements over the DER family as well as recent methods such as VR-MCL and POCL.

Citations

• InfLoRA: Interference-Free Low-Rank Adaptation for Continual Learning
• A Unified and General Framework for Continual Learning
• Online Prototype Learning for Online Continual Learning
• A Unified Continual Learning Framework with General Parameter-Efficient Tuning
• A Comprehensive Survey of Continual Learning: Theory, Method and Application
• Online Continual Learning on a Contaminated Data Stream with Blurry Task Boundaries
• Self-Sustaining Representation Expansion for Non-Exemplar Class-Incremental Learning
• An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
• The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization
• Supermasks in Superposition
• Knowledge Distillation: A Survey
• Generative Feature Replay For Class-Incremental Learning
• Dark Experience for General Continual Learning: a Strong, Simple Baseline
• Memory-Efficient Incremental Learning Through Feature Adaptation
• Compacting, Picking and Growing for Unforgetting Continual Learning
• Lifelong GAN: Continual Learning for Conditional Image Generation
• Gradient based sample selection for online continual learning
• Learning without Memorizing
• Efficient Lifelong Learning with A-GEM
• Learning to Learn without Forgetting by Maximizing Transfer and Minimizing Interference
• Memory Replay GANs: learning to generate images from new categories\n without forgetting
• End-to-End Incremental Learning
• Deep Generative Dual Memory Network for Continual Learning
• Lifelong Learning with Dynamically Expandable Networks
• Gradient Episodic Memory for Continual Learning
• Continual Learning with Deep Generative Replay
• Learning without Forgetting
• Deep Residual Learning for Image Recognition
• A Kernel Method for the Two-Sample Problem
• Random sampling with a reservoir
• A Stochastic Approximation Method

Related

• Mitigating Forgetting in Online Continual Learning via Contrasting Semantically Distinct Augmentations
• Optimizing Class Distribution in Memory for Multi-Label Online Continual Learning
• Online Continual Learning under Extreme Memory Constraints
• Sample Condensation in Online Continual Learning
• Forgetting, Ignorance or Myopia: Revisiting Key Challenges in Online Continual Learning