arXiv AI By Xiaoyan Li, Lanpei Li, Massimo Coppola, Vincenzo Lomonaco, Antonio Carta

STABLE: A Continual Learning Optimizer with Adaptive Drift Control

Read the original on arXiv AI →

STABLE is a primal‑dual optimizer that controls parameter drift to prevent catastrophic forgetting in continual learning. By enforcing a drift budget through adaptive dual variables, it guarantees bounded margin erosion for distance‑based classifiers under a local score‑stability condition. Experiments on unimodal and multimodal streams show that models trained with STABLE match or outperform specialized continual learning methods while preserving zero‑shot transfer.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.