Rehearsal-Free Modular and Compositional Continual Learning for Language Models

arXiv:2404.00790v1 Announce Type: new
Abstract: Continual learning aims at incrementally acquiring new knowledge while not forgetting existing knowledge. To overcome catastrophic forgetting, methods are either rehearsal-based, i.e., store data examples from previous tasks for data replay, or isolate parameters dedicated to each task. However, rehearsal-based methods raise privacy and memory issues, and parameter-isolation continual learning does not consider interaction between tasks, thus hindering knowledge transfer. In this work, we propose MoCL, a rehearsal-free Modular and Compositional Continual Learning framework which …

abstract arxiv catastrophic forgetting continual cs.cl cs.lg data examples free however knowledge language language models memory modular parameters privacy raise store tasks type