Overcoming Catastrophic Forgetting: A Novel Fine-Tuning Method
Keywords: lifelong learning
TL;DR: Updating knowledge by updating the most relevant parameters.
Abstract: Despite remarkable advances in Large Language Models (LLMs), a persistent challenge remains: the potential for these models to acquire erroneous or outdated information from their training data. Direct fine-tuning with data containing new knowledge can be ineffective due to conflicts between old and new knowledge. This paper proposes a novel fine-tuning paradigm called Delicate Fine-Tuning (DFT ) that leverages parametric arithmetic to pinpoint the location of knowledge and update only the minimal set of relevant parameters. Experimental results on two publicly available datasets demonstrate that our proposed DFT significantly improves the knowledge updating performance of full fine-tuning, consistently outperforming existing baselines in most cases.
Primary Area: transfer learning, meta learning, and lifelong learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 1468
Loading