Date: Wednesday, December 7th, 2022
9:00 am – 10:00 am Pacific Time
12:00 pm – 1:00 pm Eastern Time
Location: Weekly Seminar, Zoom
Title: Learning, Incentivizing Improvement, and Fairness in the Presence of Strategic Behavior
Abstract:
In this talk I will discuss a few lines of work involving learning, incentivizing improvement, and fairness in the presence of strategic behavior. First, we consider an online linear classification problem where agents arrive one by one and they wish to be classified as positive. They observe the current prediction rule and manipulate their features to get classified as positive if they can do so for a cost less than their value for being classified as positive. We show an algorithm that makes a bounded number of mistakes in presence of strategic agents for both \ell_2 and weighted \ell_1 manipulation costs. Next, we consider an offline model where agents can do both improvement and gaming modifications to their initial feature sets. The goal is to design classifiers that encourage agents to improve and become truly qualified. In the last piece of work, we consider some fairness objectives in a pure improvement setting.
Based on joint work with Hedyeh Beyhaghi, Avrim Blum, and Keziah Naggita.
Bio:
Saba Ahmadi is a postdoc at TTIC hosted by Avrim Blum. She received her Ph.D. from the University of Maryland College Park where she was advised by Samir Khuller. During her Ph.D., she visited Northwestern University for 2 years. She is interested in the social aspects of computing, learning theory, and combinatorial optimization..