Speaker
Amit Deshpande (Microsoft Research)
Date & Time
Fri, 07 February 2025, 11:00 to 12:30
Venue
Madhava Lecture Hall
Resources
Abstract
Modeling preference data over multiple choices is useful in voting, search, recommendations, economics, and more recently, in the feedback used to improve AI models. In this talk, I will give a quick overview of some probabilistic choice models such as Bradley-Terry and Plackett-Luce. We will rederive their nice mathematical properties (e.g., IIA or Independence of Irrelevant Alternatives), highlight a few of their limitations (e.g., IIA), and explore newer extensions tailored to fair ranking, AI alignment and finetuning.
Zoom link: https://icts-res-in.zoom.us/j/91503040475?pwd=zNTGnlTnSCam40q7bjVNDFMBGu7aUa.1
Meeting ID: 915 0304 0475
Passcode: 070809