Transferability Between Regression Tasks

Cuong Ngoc Nguyen, Phong Tran The, Lam Si Tung Ho, Vu C. Dinh, Anh Tuan Tran, Tal Hassner, Cuong V Nguyen

Keywords: Transferability estimation, Transfer learning

Abstract: We consider the problem of estimating how well deep neural network regression models would transfer from source to target tasks. We focus on regression tasks, which received little previous attention, and develop novel transferability estimation methods that are simple, computationally efficient, yet effective and theoretically grounded. We propose two families of transferability estimators, both of which utilize the mean squared error of a regularized linear regression model to estimate the transferability. We prove novel theoretical bounds connecting our methods with the expected risk of the optimal target models obtained from the actual transfer learning process. We test our methods extensively in various challenging, practical scenarios and show they significantly outperform existing state-of-the-art regression task transferability estimators, in both accuracy and efficiency.

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.

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: Yes

Please Choose The Closest Area That Your Submission Falls Into: Deep Learning and representational learning