Fine-tuning can cripple foundation models; preserving features may be the solution

Jishnu Mukhoti, Yarin Gal, Philip Torr, Puneet K. Dokania

Keywords: concept forgetting, foundation model

Abstract: Pre-trained foundation models, due to their enormous capacity and their training using vast amounts of data can store knowledge about many real-world concepts. To further improve performance on downstream tasks, these models can be fine-tuned on task specific datasets. While various fine-tuning methods have been devised and have been shown to be highly effective, we observe that a fine-tuned model's ability to recognize concepts on tasks different from the downstream one is reduced significantly compared to its pre-trained counterpart. This is clearly undesirable as a huge amount of time and money went into learning those very concepts in the first place. We call this undesirable phenomenon "concept forgetting'' and via experiments show that most end-to-end fine-tuning approaches suffer heavily from this side effect. To this end, we also propose a rather simple fix to this problem by designing a method called LDIFS (short for $\ell_2$ distance in feature space) that simply preserves the features of the original foundation model during fine-tuning. We show that LDIFS significantly reduces concept forgetting without having noticeable impact on the downstream task performance.

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/2024/AuthorGuide.

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: 5434