Fair and Efficient Online Decision Making
Yusuf Hakan Kalaycı
University of Southern California, Los Angeles
Abstract: As algorithms increasingly automate critical decisions—from AI-powered text generation to democratic representation—it is more important than ever to design decision-making systems that are both efficient and fair. This talk addresses a fundamental challenge in online decision making: how can we decide efficiently and fairly when information arrives sequentially and the future is uncertain? To illustrate my research in this direction, I will present two recent examples. First, on the efficiency front, I show how optimal stopping theory, specifically through the lens of Pandora’s Box, can dramatically improve the efficiency of large language model inference by learning when to stop generating tokens. Second, on the fairness front, I demonstrate how principles from computational social choice can make the selection of permanent citizens’ assemblies more representative over time, ensuring fair representation even for smaller groups as decisions accumulate. Together, these examples showcase how theoretical foundations in online algorithms can address practical challenges in modern AI systems and democratic institutions.
Biography: Yusuf Hakan Kalayci is a Ph.D. candidate in Computer Science at the University of Southern California, where he researches optimal stopping theory and computational social choice with applications to large language models. He earned his B.Sc. in Computer Engineering and Mathematics with High Honors from Boğaziçi University. His research has been published in leading conferences including STOC, ITCS, ICALP, AAAI, and AAMAS, and he has served as a reviewer and/or program committee member for STOC, ESA, and AAAI.
Date: March 2, 2026 Monday @ 10:00
Place: Zoom
This is an online seminar. To request event details please send a message to department.