Abstract: Many instances of algorithmic bias are caused by distribution shifts, but most algorithmic fairness practices were not developed with distribution shifts in mind. This discrepancy between the causes of algorithmic biases and the premise of prior work on algorithmic fairness leads us to study whether enforcing fairness mitigates algorithmic biases caused by distribution shifts. On one hand, we show that there are scenarios in which enforcing fairness does not improve model performance. In fact, it may even harm performance. On the other hand, we derive sufficient conditions under which enforcing group and individual fairness successfully mitigates algorithmic biases due to distribution shifts.


Yuekai Sun is an assistant professor in the statistics department at the University of Michigan. His research is guided by the statistical and computational challenges in machine learning. Some topics of recent interest are algorithmic fairness, transfer learning, and federated learning.

https://yuekai.github.io/