Towards Explaining Distribution Shifts
This work addresses the need for interpretable explanations of distribution shifts in machine learning, which is critical for human operators to understand and mitigate shifts in operating environments, though it is incremental as it builds on prior detection-focused methods.
The paper tackles the problem of explaining distribution shifts to aid manual mitigation tasks by proposing interpretable transportation maps derived from a relaxation of optimal transport, and demonstrates on real-world datasets that these mappings provide a better balance between detail and interpretability than baselines.
A distribution shift can have fundamental consequences such as signaling a change in the operating environment or significantly reducing the accuracy of downstream models. Thus, understanding distribution shifts is critical for examining and hopefully mitigating the effect of such a shift. Most prior work focuses on merely detecting if a shift has occurred and assumes any detected shift can be understood and handled appropriately by a human operator. We hope to aid in these manual mitigation tasks by explaining the distribution shift using interpretable transportation maps from the original distribution to the shifted one. We derive our interpretable mappings from a relaxation of optimal transport, where the candidate mappings are restricted to a set of interpretable mappings. We then inspect multiple quintessential use-cases of distribution shift in real-world tabular, text, and image datasets to showcase how our explanatory mappings provide a better balance between detail and interpretability than baseline explanations by both visual inspection and our PercentExplained metric.