Finding Structure in Continual Learning

2026-02-04

AI Summary

This work addresses the stability-plasticity dilemma in continual learning, where acquiring new knowledge often leads to catastrophic forgetting of previously learned information. The authors propose a novel approach by introducing the Douglas–Rachford splitting method to decouple the original optimization problem into two sub-objectives: one enhancing plasticity by facilitating learning on new tasks, and the other promoting stability by preserving knowledge from prior tasks. These sub-problems are solved iteratively via proximal operators that converge to a consensus solution. Notably, the method operates without external replay buffers or complex regularization schemes, thereby circumventing gradient conflicts commonly arising from multi-loss weighting strategies. By avoiding reliance on auxiliary modules, the proposed framework achieves substantial improvements in both model performance and training stability across sequential learning scenarios.

Technology Category

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

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Learning from a stream of tasks usually pits plasticity against stability: acquiring new knowledge often causes catastrophic forgetting of past information. Most methods address this by summing competing loss terms, creating gradient conflicts that are managed with complex and often inefficient strategies such as external memory replay or parameter regularization. We propose a reformulation of the continual learning objective using Douglas-Rachford Splitting (DRS). This reframes the learning process not as a direct trade-off, but as a negotiation between two decoupled objectives: one promoting plasticity for new tasks and the other enforcing stability of old knowledge. By iteratively finding a consensus through their proximal operators, DRS provides a more principled and stable learning dynamic. Our approach achieves an efficient balance between stability and plasticity without the need for auxiliary modules or complex add-ons, providing a simpler yet more powerful paradigm for continual learning systems.

Problem

• continual learning
• catastrophic forgetting
• stability-plasticity dilemma
• task stream

Innovation

• Continual Learning
• Douglas-Rachford Splitting
• Plasticity-Stability Trade-off
• Proximal Operators
• Gradient Conflict