TY - GEN
T1 - Fairness violations and mitigation under covariate shift
AU - Singh, Harvineet
AU - Singh, Rina
AU - Mhasawade, Vishwali
AU - Chunara, Rumi
N1 - Publisher Copyright:
© 2021 ACM.
PY - 2021/3/3
Y1 - 2021/3/3
N2 - 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.
AB - 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.
KW - Algorithmic fairness
KW - Causal inference
KW - Covariate shift
KW - Domain adaptation
UR - http://www.scopus.com/inward/record.url?scp=85102613983&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85102613983&partnerID=8YFLogxK
U2 - 10.1145/3442188.3445865
DO - 10.1145/3442188.3445865
M3 - Conference contribution
AN - SCOPUS:85102613983
T3 - FAccT 2021 - Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency
SP - 3
EP - 13
BT - FAccT 2021 - Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency
PB - Association for Computing Machinery, Inc
T2 - 4th ACM Conference on Fairness, Accountability, and Transparency, FAccT 2021
Y2 - 3 March 2021 through 10 March 2021
ER -