ISE Seminar Series - Santiago Segarra

Title: Task-Informed and Controllable Generative Flows for Inverse Problems

Abstract:

Generative flow models provide powerful priors for inverse problems, but the standard formulation is rigid: the model starts from a task-agnostic source, follows fixed sampling dynamics, and returns a single type of estimate. In this talk, we discuss three ways to make generative flows more task-informed and controllable. First, we introduce prior-informed flow matching, where the source distribution is conditioned on the available observations rather than chosen as generic Gaussian noise. In graph reconstruction, a simple initial estimator provides the starting point, while the learned flow captures global structural dependencies missed by edge-wise predictors. Second, we show how to steer the sampling trajectory when structural side information is available, such as degree bounds or triangle counts: Lagrange multipliers that evolve along each trajectory respond to predicted constraint violations and guide the subsequent steps, raising feasibility without retraining or tuning a penalty per constraint. Third, we present flow-map denoisers, which learn finite-time flow operators rather than only instantaneous velocity fields. A lookahead parameter then selects an estimator between the conditional mean and a perceptual sampler, so a single model traverses the distortion-perception trade-off. Together, these ideas suggest a broader view of generative models for inverse problems: we can design where the trajectory starts, how it is steered, and which point along it best serves the task at hand.

Bio: 
Santiago Segarra received the B.Sc. degree in Industrial Engineering with highest honors (Valedictorian) from the Instituto Tecnológico de Buenos Aires (ITBA), Argentina, in 2011, the M.Sc. in Electrical Engineering from the University of Pennsylvania (Penn), Philadelphia, in 2014 and the Ph.D. degree in Electrical and Systems Engineering from Penn in 2016. From September 2016 to June 2018, he was a postdoctoral research associate with the Institute for Data, Systems, and Society at the Massachusetts Institute of Technology. He joined Rice University in 2018 as an Assistant Professor and, since July 2024, Dr. Segarra is a W. M. Rice Trustee Associate Professor in the Department of Electrical and Computer Engineering at Rice University. He also holds courtesy appointments in the Departments of Computer Science and Statistics. His research interests include network theory, data analysis, machine learning, and graph signal processing. Dr. Segarra received the 2011 Outstanding Graduate Award granted by the National Academy of Engineering of Argentina, the 2017 Penn's Joseph and Rosaline Wolf Award for Best Doctoral Dissertation in Electrical and Systems Engineering, the 2020 IEEE Signal Processing Society Young Author Best Paper Award, the 2021 Rice's School of Engineering Research + Teaching Excellence Award, three early career awards (NSF CAREER, ARO ECP, and ARI Early Career), and five best conference paper awards. He has dedicated part of his time to consulting for Microsoft Research and, in 2024, co-founded Aqmen AI where he currently serves as CTO.