Ali Tajer, "Causal Representation Learning"

Abstract: Building reliable world models for physical systems is a central challenge in modern AI, with broad implications for scientific discovery, robotics, autonomy, and control. Unlike purely statistical prediction, world models for physical environments must capture the underlying causal structure of the system: which latent factors correspond to genuine physical mechanisms, how these factors evolve under interventions, and how they govern future observations across changing environments. This motivates a fundamental question: how can an agent learn representations that are not merely predictive, but causally meaningful and therefore suitable for reasoning, planning, and adaptation in the physical world? In this talk, we will discuss causal representation learning (CRL) as a principled foundation for building world models of physical systems. CRL aims to uncover latent variables that correspond to the system’s underlying generative factors and to organize them in a way that reflects causal relationships, invariances, and intervention effects. Such representations can enable models that extrapolate more reliably beyond the training regime, support counterfactual and interventional reasoning, and provide a pathway toward more interpretable and transferable decision-making. We will present our perspective on why causality is essential for world modeling, outline key challenges in learning causal latent structure from high-dimensional observations, and highlight emerging directions for integrating CRL with dynamics, control, and scientific machine learning. Bio: Ali Tajer received a B.Sc. and an M.Sc. degree in Electrical Engineering from Sharif University of Technology, an M.A. in Statistics, and a Ph.D. in Electrical Engineering from Columbia University. During 2010-2012, he was a Postdoctoral Research Associate at Princeton University. He is currently a Professor of Electrical, Computer, and Systems Engineering and the Associate Dean for Academic Affairs in th