Quality-Diversity Transfer Learning (QDTL)

Samridhi Vaid, Simon Urschel, David Mark Olson, Matthew Guzdial

Keywords: Transfer Learning, Medical Prediction, Quality Diversity

Abstract: Deep learning has had much success on challenging problems with large datasets. However, it struggles in cases with limited training data. Transfer learning represents a class of approaches for transferring knowledge from large source datasets to smaller target datasets. But many transfer learning approaches have constraints in terms of dataset size and similarity of output features. In this paper, we introduce Quality-Diversity Transfer Learning (QDTL), a novel transfer learning approach based on neuroevolution for dealing with very small dataset problems with distinct output features. We demonstrate the success of QDTL on two medical prediction problems, outperforming standard transfer learning baselines.

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: 5921