Date: March 6, 2026
Speaker: Juan Carlos Perdomo
Title: The Relative Value of Prediction
Abstract: Predictive algorithms are increasingly used to guide the allocation of scarce resources—from deciding which students receive tutoring in Wisconsin to targeting cash transfers in sub-Saharan Africa. In these contexts, predictions are only a means to an end: they help planners make better decisions with the ultimate goal of improving social welfare, such as increasing graduation rates or reducing poverty. Viewed from this perspective, it’s clear that prediction is only one part of a rich design space. There are many other ways to improve welfare beyond increasing predictive accuracy such as expanding access to the underlying good or in increasing treatment effectiveness. Given this broader set of design decisions, when are investments in prediction truly worth it? In this talk I will discuss a line of work that provides empirical and mathematical tools for answering this question.
Based on joint work with Emily Aiken, Christoph Kern, and Unai Fischer-Abaigar.
Bio: Juan Carlos Perdomo is a postdoctoral fellow at MIT and will join New York University as an assistant professor of computer science and data science in the fall of 2026. His research focuses on the foundations of machine learning systems that make predictions or decisions about people. He earned his Ph.D. in Electrical Engineering and Computer Science from UC Berkeley and was previously CRCS Postdoctoral Fellow at Harvard. His work was recently recognized with an Oustanding Paper Award at ICML.