Saved in:
Bibliographic Details
Main Author: Khilar, Snigdha Chandan
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2606.01863
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917553677271040
author Khilar, Snigdha Chandan
author_facet Khilar, Snigdha Chandan
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2606_01863
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Continual Learning as a Multiphase Moving-Boundary Problem
Khilar, Snigdha Chandan
Machine Learning
Mathematical Physics
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.
title Continual Learning as a Multiphase Moving-Boundary Problem
topic Machine Learning
Mathematical Physics
url https://arxiv.org/abs/2606.01863