Babak Salimi (UCSD): Certifying the Fairness of Predictive Models in the Face of Selection Bias

The Department of Computer Science, UC Irvine

WELCOMES

Prof. Babak Salimi

UCSD

Hosts: Prof. Chen Li

Certifying the Fairness of Predictive Models in the Face of Selection Bias

Abstract: The widespread use of data-driven algorithmic decision making in crucial areas such as hiring, loan assessments, medical diagnoses, and pretrial release has raised questions about the accuracy and fairness of these algorithms. Selection bias, a prevalent data quality issue in sensitive domains, is a major obstacle to creating fair predictive models. Most existing fair predictive modeling approaches are unable to address selection bias. To overcome this challenge, we introduce a new framework called CRAB that leverages principles of data management and query answering from inconsistent and incomplete databases to produce certifiably fair predictive models.

In this talk, we will delve into the concept of consistent range approximation, which plays a critical role in approximating the fairness of predictive models on a target population using biased data. We will also discuss the difficulties in achieving consistent range approximation when limited or no external data is available. With the help of our framework, CRAB, we can train predictive models that are certifiably fair on the target population, even in the presence of selection bias. This talk will provide valuable insights for those working in data management, ML, and responsible data science and emphasize the importance of addressing selection bias in algorithmic decision making.

Bio: Babak Salimi is an Assistant Professor in the HDSI department at UC San Diego. Prior to this, he was a postdoctoral researcher in the Computer Science and Engineering Department at the University of Washington, where he collaborated with Prof. Dan Suciu and the database group. Salimi received his Ph.D. from Carleton University’s School of Computer Science, where he was advised by Prof. Leopoldo Bertossi. His research focuses on responsible data management and causal inference, including algorithmic fairness and transparency. He has made several significant contributions to the understanding of responsible data management and analysis, including explainability, fairness, reliability, and robustness. Salimi also has a strong interest in database theory and data management. His research achievements have been acknowledged with awards such as the Postdoc Research Award at the University of Washington, the Best Demonstration Paper Award at VLDB 2018, the Best Paper Award at SIGMOD 2019, and the Research Highlight Award at SIGMOD 2020.