The Evolution of Out-of-Distribution Robustness Throughout Fine-Tuning
This addresses the problem of improving out-of-distribution performance for machine learning practitioners, but it is incremental as it builds on existing empirical trends.
The paper investigates how fine-tuning affects out-of-distribution robustness, finding that models pre-trained on larger datasets show temporary effective robustness that disappears at convergence, and identifies data properties like size and diversity that enhance it, with such models correctly classifying 10% of examples missed by others.
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.