Assaying Out-Of-Distribution Generalization in Transfer Learning

TL;DR

This paper evaluates various methods for out-of-distribution generalization in transfer learning by conducting extensive experiments on a large collection of datasets, revealing complex relationships between in- and out-of-distribution performance.

Contribution

It provides a unified empirical framework for comparing OOD robustness measures and offers practical recommendations based on extensive dataset and model evaluations.

Findings

In- and out-of-distribution accuracies often increase together.

The relationship between in- and OOD performance is dataset-dependent.

More nuanced than previous smaller-scale studies.

Abstract

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.