Go to ICLR 2025 Conference homepageReassessing the Validity of Spurious Correlations Benchmarks
Keywords: benchmarking, evaluation, spurious correlations
TL;DR: We systematically explore disagreement between spurious correlations benchmarks and examine their validity, finding that certain benchmarks are not meaningful evaluations of spurious correlations mitigation method performance.
Abstract: Neural networks can fail when the data contains spurious correlations, i.e. associations in the training data that fail to generalize to new distributions. To understand this phenomenon, often referred to as subpopulation shift or shortcut learning, researchers have proposed numerous group-annotated spurious correlations benchmarks upon which to evaluate mitigation methods. However, we observe that these benchmarks exhibit substantial disagreement, with the best methods on one benchmark performing poorly on another. We explore this disagreement, and examine benchmark validity by defining three desiderata that a benchmark should satisfy in order to meaningfully evaluate methods. Our results have implications for both benchmarks and mitigations: we find that certain group-annotated benchmarks are not meaningful measures of method performance, and that several methods are not sufficiently robust for widespread use. We present a simple recipe for practitioners to choose methods using the _most similar_ benchmark to their given problem.
Primary Area: datasets and benchmarks
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/2025/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: 9851
Loading