Dataset Interfaces: Diagnosing Model Failures Using Controllable Counterfactual Generation
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 notio...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Published: |
Massachusetts Institute of Technology
2024
|
| Online Access: | https://hdl.handle.net/1721.1/156303 |
| _version_ | 1826215147173052416 |
|---|---|
| author | Vendrow, Joshua L. |
| author2 | Mądry, Aleksander |
| author_facet | Mądry, Aleksander Vendrow, Joshua L. |
| author_sort | Vendrow, Joshua L. |
| collection | MIT |
| description | 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 dataset interface 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. |
| first_indexed | 2024-09-23T16:17:19Z |
| format | Thesis |
| id | mit-1721.1/156303 |
| institution | Massachusetts Institute of Technology |
| last_indexed | 2024-09-23T16:17:19Z |
| publishDate | 2024 |
| publisher | Massachusetts Institute of Technology |
| record_format | dspace |
| spelling | mit-1721.1/1563032024-08-22T03:03:35Z Dataset Interfaces: Diagnosing Model Failures Using Controllable Counterfactual Generation Vendrow, Joshua L. Mądry, Aleksander Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science 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 dataset interface 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. S.M. 2024-08-21T18:55:14Z 2024-08-21T18:55:14Z 2024-05 2024-07-10T13:00:00.670Z Thesis https://hdl.handle.net/1721.1/156303 In Copyright - Educational Use Permitted Copyright retained by author(s) https://rightsstatements.org/page/InC-EDU/1.0/ application/pdf Massachusetts Institute of Technology |
| spellingShingle | Vendrow, Joshua L. Dataset Interfaces: Diagnosing Model Failures Using Controllable Counterfactual Generation |
| title | Dataset Interfaces: Diagnosing Model Failures Using Controllable Counterfactual Generation |
| title_full | Dataset Interfaces: Diagnosing Model Failures Using Controllable Counterfactual Generation |
| title_fullStr | Dataset Interfaces: Diagnosing Model Failures Using Controllable Counterfactual Generation |
| title_full_unstemmed | Dataset Interfaces: Diagnosing Model Failures Using Controllable Counterfactual Generation |
| title_short | Dataset Interfaces: Diagnosing Model Failures Using Controllable Counterfactual Generation |
| title_sort | dataset interfaces diagnosing model failures using controllable counterfactual generation |
| url | https://hdl.handle.net/1721.1/156303 |
| work_keys_str_mv | AT vendrowjoshual datasetinterfacesdiagnosingmodelfailuresusingcontrollablecounterfactualgeneration |