Data-dependent and Oracle Bounds on Forgetting in Continual Learning

2024-06-13

arXiv.org

AI Summary

This study addresses the quantification and control of catastrophic forgetting of old-task knowledge in continual learning. We first propose data-dependent and oracle-constrained upper bounds on forgetting—agnostic to specific models or algorithms—and extend them to the Gibbs posterior framework. Leveraging information-theoretic analysis and generalization error bounds, we derive an optimization-friendly forgetting suppression algorithm. Theoretical bounds are empirically validated as tight across multiple benchmarks. Experiments demonstrate that our algorithm significantly reduces forgetting while preserving cross-task positive forward transfer. Our core contributions are: (1) establishing the first universal, computable forgetting bound that jointly incorporates data structure and prior information; and (2) enabling theory-driven, controllable continual learning with explicit forgetting regulation.

Technology Category

• Machine Learning: Life-Long and Continual Learning
• Search and Optimization: Learning to Search
• Constraint Satisfaction and Optimization: Constraint Learning and Acquisition

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics

Abstract

In continual learning, knowledge must be preserved and re-used between tasks, maintaining good transfer to future tasks and minimizing forgetting of previously learned ones. While several practical algorithms have been devised for this setting, there have been few theoretical works aiming to quantify and bound the degree of Forgetting in general settings. We provide both data-dependent and oracle upper bounds that apply regardless of model and algorithm choice, as well as bounds for Gibbs posteriors. We derive an algorithm based on our bounds and demonstrate empirically that our approach yields tight bounds on forgetting for several continual learning problems.

Problem

• Continuous Learning
• Knowledge Retention
• Forgetting Mitigation

Innovation

• Quantifiable Forgetting
• Continuous Learning
• Assessment Tool