Two Studies in AI & Experimentation: Model Behavior Alignment & Network Interference Effects – Prof. Mohsen Bayati

Join us for the Data Sciences Speaker Series with Prof. Mohsen Bayati, Professor at Stanford University.  This talk is co-sponsored by the Data Sciences Institute and the Centre for Analytics and Artificial Intelligence Engineering (CARTE), University of Toronto.

• Date: May 26, 2025
• Time: 11:00 a.m. – 12:00 p.m.
• Format: In-person
• Location: Data Sciences Institute, 10th floor Seminar Room, 700 University Avenue, Toronto 

Two Studies in AI & Experimentation: Model Behavior Alignment & Network Interference Effects
This talk presents research on two problems in applied AI and experimental design. Examples of AI alignment challenges from healthcare applications include: shape-constrained learning in clinical settings and language model deployment in pharmacy systems. These cases demonstrate how models can successfully optimize their training objectives and achieve excellent out-of-sample performance, yet still fail to meet critical requirements set by clinical professionals who are the end users. This misalignment persists in non-generative settings even with abundant data, and the talk will present a conformal alignment framework as a unified solution to address these challenges. A new statistical physics-inspired framework for analyzing network interference in experimental settings will be introduced. Unlike traditional approaches that primarily focus on equilibrium outcomes, this framework leverages the rich information contained in the dynamic propagation of treatment effects through networks. The causal message-passing methodology captures a complex propagation process, enabling reliable estimation of treatment effects without requiring knowledge of the underlying network structure or specific interaction mechanisms.

Biography
Prof. Bayati’s research focuses on data-driven decision-making and experiment design, particularly as they intersect with healthcare and e-commerce. He utilizes tools from contextual multi-armed bandits, graphical models, message-passing algorithms, and high-dimensional statistics. Mohsen received a BS in Mathematics from Sharif University of Technology and a PhD in Electrical Engineering from Stanford University. He then worked as a postdoctoral researcher at Microsoft Research and Stanford University. His work was awarded the INFORMS Healthcare Applications Society’s Best Paper (Pierskalla) Award in 2014 and 2016, the INFORMS Applied Probability Society’s Best Paper Award in 2015, and the National Science Foundation CAREER Award.