Assaying Out-Of-Distribution Generalization in Transfer Learning

Since out-of-distribution generalization is a generally ill-posed problem, various proxy targets (e.g., calibration, adversarial robustness, algorithmic corruptions, invariance across shifts) were studied across different research programs resulting in different recommendations. While sharing the same aspirational goal, these approaches have never been tested under the same experimental conditions on real data. In this paper, we take a unified view of previous work, highlighting message discrepancies that we address empirically, and providing recommendations on how to measure the robustness of a model and how to improve it. To this end, we collect 172 publicly available dataset pairs for training and out-of-distribution evaluation of accuracy, calibration error, adversarial attacks, environment invariance, and synthetic corruptions. We fine-tune over 31k networks, from nine different architectures in the many- and few-shot setting. Our findings confirm that in- and out-of-distribution accuracies tend to increase jointly, but show that their relation is largely dataset-dependent, and in general more nuanced and more complex than posited by previous, smaller scale studies. and out-of-distribution (OOD) dataset pairs, fine-tuning and evaluating over 31k models to gain a broader insight in the sometimes contradicting statements on OOD robustness in previous research. We organize our study around two key questions: (1) What are good proxy measures of OOD robustness when having access to a single dataset? (2) How do architecture choices and fine-tuning strategies affect robustness? We plan to publish the code with the camera-ready version of the paper.