The Evolution of Out-of-Distribution Robustness Throughout Fine-Tuning

TL;DR

This paper investigates how models' robustness to out-of-distribution data evolves during fine-tuning, revealing that larger, more diverse datasets influence robustness, which can be leveraged to enhance model performance on OOD data.

Contribution

It provides a comprehensive empirical analysis of effective robustness during fine-tuning, highlighting the impact of dataset properties and proposing strategies to scale robustness for better OOD performance.

Findings

Models pre-trained on larger datasets exhibit effective robustness that vanishes at convergence.

Effective robustness increases with dataset size, diversity, and example difficulty.

Models with effective robustness correctly classify 10% of examples others fail to classify.

Abstract

Although machine learning models typically experience a drop in performance on out-of-distribution data, accuracies on in- versus out-of-distribution data are widely observed to follow a single linear trend when evaluated across a testbed of models. Models that are more accurate on the out-of-distribution data relative to this baseline exhibit "effective robustness" and are exceedingly rare. Identifying such models, and understanding their properties, is key to improving out-of-distribution performance. We conduct a thorough empirical investigation of effective robustness during fine-tuning and surprisingly find that models pre-trained on larger datasets exhibit effective robustness during training that vanishes at convergence. We study how properties of the data influence effective robustness, and we show that it increases with the larger size, more diversity, and higher example difficulty of the dataset. We also find that models that display effective robustness are able to correctly classify 10% of the examples that no other current testbed model gets correct. Finally, we discuss several strategies for scaling effective robustness to the high-accuracy regime to improve the out-of-distribution accuracy of state-of-the-art models.