Timezone: »

 
Dataset Interfaces: Diagnosing Model Failures Using Controllable Counterfactual Generation
Joshua Vendrow · Saachi Jain · Logan Engstrom · Aleksander Madry

@
in DMLR Workshop: Data-centric Machine Learning Research »

Distribution shift is a major source of failure for machine learning models. However, evaluating model reliability under distribution shift can be challenging, especially since it may be difficult to acquire counterfactual examples that exhibit a specified shift. In this work, we introduce the notion of a dataset interface: a framework that, given an input dataset and a user-specified shift, returns instances from that input distribution that exhibit the desired shift. We study a number of natural implementations for such an interface, and find that they often introduce confounding shifts that complicate model evaluation. Motivated by this, we propose a new implementation that leverages Textual Inversion to tailor generation to the input distribution. We then demonstrate how applying this dataset interface to the ImageNet dataset enables studying model behavior across a diverse array of distribution shifts, including variations in background, lighting, and attributes of the objects.

Author Information

Joshua Vendrow (Massachusetts Institute of Technology)
Saachi Jain (Massachusetts Institute of Technology)
Logan Engstrom (MIT)
Aleksander Madry (MIT)

More from the Same Authors