Forgetting of task-specific knowledge in model merging-based continual learning

Timm Hess; Gido M van de Ven; Tinne Tuytelaars

Forgetting of task-specific knowledge in model merging-based continual learning

Computer Vision and Pattern Recognition 2025-08-01 v1

Authors: Timm Hess , Gido M van de Ven , Tinne Tuytelaars

Abstract

This paper investigates the linear merging of models in the context of continual learning (CL). Using controlled visual cues in computer vision experiments, we demonstrate that merging largely preserves or enhances shared knowledge, while unshared task-specific knowledge rapidly degrades. We further find that merging models from an incremental training process consistently outperforms merging models trained in parallel.

Keywords

continual learning multi-task learning multi-view learning

Cite

@article{arxiv.2507.23311,
  title  = {Forgetting of task-specific knowledge in model merging-based continual learning},
  author = {Timm Hess and Gido M van de Ven and Tinne Tuytelaars},
  journal= {arXiv preprint arXiv:2507.23311},
  year   = {2025}
}

Related papers

View all related →

Computer Vision and Pattern Recognition · Computer Science

Continual Learning in Vision-Language Models via Aligned Model Merging

Ghada Sokar, Gintare Karolina Dziugaite, Anurag Arnab, Ahmet Iscen +2