Continual Learning as a Multiphase Moving-Boundary Problem
2026-06-01 • Machine Learning
Machine Learning
AI summaryⓘ
The authors present Stefan-CL, a new way to help AI systems learn new tasks without forgetting old ones. They use ideas from physics, imagining what the AI knows well as a solid and free space to learn as a liquid that changes as learning happens. Their method keeps the solid part protected, reducing forgetting almost to zero, without needing to save old data. This approach balances remembering and learning in a novel, physics-inspired way.
Continual learningStability-plasticity dilemmaCatastrophic forgettingConsolidated knowledgeLatent heatNetwork capacityPhysics-inspired modelsMemory retention
Authors
Snigdha Chandan Khilar
Abstract
Continual learning struggles to balance retaining past knowledge with absorbing new tasks. Stefan-CL elegantly resolves this stability-plasticity dilemma through the physics of melting. It frames consolidated knowledge as a protected "solid" and unused capacity as an adaptable "liquid." As the network learns, this boundary expands, governed by a "latent heat" tuning dial. By mathematically freezing the learned interior, Stefan-CL cuts forgetting to near zero, matching memory-heavy baselines without storing raw data, forging a beautiful, physics-grounded path for AI.