Towards More Realistic and Fairer Strategic Classification

Juba Ziani
Georgia Tech, Industrial and Systems Engineering

Time: 2024-11-13, 12:00 - 13:00 ET
Location: Rice 540 and Zoom

Abstract In Strategic Classification, agents facing high-stakes decision settings may attempt to game the decision to increase their chance of obtaining a more favorable decision. The general model is the following: agents face a classifier and have a cost to modify their features in order to improve their chances of passing the classifier. The learner aims to compensate for this strategic behavior by making positive classification outcomes more difficult to obtain.

The standard setting of strategic classification, however, relies on strong and unrealistic assumptions: agents have a complete understanding of the deployed decision rule or classifier; strategic behavior is usually seen as ``gaming’’, not as potentially a way to improve individuals truly; decisions are made in an offline and static fashion; and the goal is generally loss minimization without considerations of fairness nor social good.

The current talk aims to further push the understanding of strategic classification by challenging these assumptions and modeling choices. While the talk will aim to cover several of the ways in which one can relax the traditional assumptions of strategic classification, the main focus will be on i) the role of “incomplete information and information asymmetries” in strategic classification and ii) on how this impacts fairness and social welfare across populations.

Bio Juba Ziani is an Assistant Professor in the Industrial and Systems Engineering (ISyE) Department at Georgia Tech and a member of the new NSF AI Research Institute for Advances in Optimization (AI4OPT) institute. Prof. Ziani’s research spans multiple dimensions at the intersection of computer science, operations research, and economics. In particular, Prof. Ziani works on market design, learning theory (with a focus on strategic environments), data privacy (with a focus on differential privacy), and fairness (both in an algorithmic and in a game-theoretic/socio-economic sense). Prof. Ziani’s research has been recognized through an NSF CAREER Award in CISE, Human-Centered Computation (2024), as well as a Best Paper Award at SATML 2023.

Prior to Georgia Tech, Prof. Ziani was a Warren Center Postdoctoral Fellow at the University of Pennsylvania, hosted by Sampath Kannan, Michael Kearns, and Aaron Roth; and a Ph.D. student at Caltech in Computer Science with Katrina Ligett and Adam Wierman.