ERM++: An Improved Baseline for Domain Generalization

Piotr Teterwak, Kuniaki Saito, Theodoros Tsiligkaridis, Kate Saenko, Bryan A. Plummer

Keywords: Domain Generalization, Multi-Source Domain Generalization.

TL;DR: We add to ERM as a baseline for Domain Generalization, and call the method ERM++.

Abstract: Multi-source Domain Generalization (DG) measures a classifier's ability to generalize to new distributions of data it was not trained on, given several training domains. While several multi-source DG methods have been proposed, they incur additional complexity during training by using domain labels. Recent work has shown that a well-tuned Empirical Risk Minimization (ERM) training procedure, that is simply minimizing the empirical risk on the source domains, can outperform most existing DG methods. ERM has achieved such strong results while only tuning hyper-parameters such as learning rate, weight decay, and batch size. This paper aims to understand how we can push ERM as a baseline for DG further, thereby providing a stronger baseline for which to benchmark new methods. We call the resulting improved baseline ERM++, and it consists of better utilization of training data, model parameter selection, and weight-space regularization. ERM++ significantly improves the performance of DG on five multi-source datasets by over 5% compared to standard ERM using ResNet-50, and beats state-of-the-art despite being less computationally expensive. We also demonstrate the efficacy of ERM++ on the WILDS-FMOW dataset, a challenging DG benchmark. Finally, we show that with a CLIP-pretrained ViT-B/16, ERM++ outperforms ERM by over 10%, allowing one to take advantage of the stronger pre-training effectively. We will release code upon acceptance.

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: 5994