Fairness violations and mitigation under covariate shift

Harvineet Singh, Rina Singh, Vishwali Mhasawade, Rumi Chunara

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

We study the problem of learning fair prediction models for unseen test sets distributed differently from the train set. Stability against changes in data distribution is an important mandate for responsible deployment of models. The domain adaptation literature addresses this concern, albeit with the notion of stability limited to that of prediction accuracy. We identify sufficient conditions under which stable models, both in terms of prediction accuracy and fairness, can be learned. Using the causal graph describing the data and the anticipated shifts, we specify an approach based on feature selection that exploits conditional independencies in the data to estimate accuracy and fairness metrics for the test set. We show that for specific fairness definitions, the resulting model satisfies a form of worst-case optimality. In context of a healthcare task, we illustrate the advantages of the approach in making more equitable decisions.

Original languageEnglish (US)
Title of host publicationFAccT 2021 - Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency
PublisherAssociation for Computing Machinery, Inc
Pages3-13
Number of pages11
ISBN (Electronic)9781450383097
DOIs
StatePublished - Mar 3 2021
Event4th ACM Conference on Fairness, Accountability, and Transparency, FAccT 2021 - Virtual, Online, Canada
Duration: Mar 3 2021Mar 10 2021

Publication series

NameFAccT 2021 - Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency

Conference

Conference4th ACM Conference on Fairness, Accountability, and Transparency, FAccT 2021
Country/TerritoryCanada
CityVirtual, Online
Period3/3/213/10/21

Keywords

  • Algorithmic fairness
  • Causal inference
  • Covariate shift
  • Domain adaptation

ASJC Scopus subject areas

  • Business, Management and Accounting(all)

Fingerprint

Dive into the research topics of 'Fairness violations and mitigation under covariate shift'. Together they form a unique fingerprint.

Cite this