Introduction to Causal Representation Learning
Modern AI systems have become remarkably good at recognizing patterns, however statistical pattern matching alone is not enough to build systems that are robust, generalizable, or capable of reasoning about cause and effect. In this talk, we explore how representation learning, which aims to equip machines with structured internal models of their inputs, can be enriched with causal structure to produce more reliable and interpretable AI. We will begin with an introduction to representation learning and its limitations, then examine how incorporating the principle of Independent Causal Mechanisms changes the way models represent and respond to their environment. Drawing on landmark as well as on recent papers in the field, we will trace how causal representation learning has evolved from a theoretical idea into a practical framework shaping modern machine learning research. We will also touch on the recently emerging world model literature, which envisions AI systems that go beyond reactive prediction to simulate and reason about how the world changes. The talk concludes with a forward-looking discussion of where this field is headed and what it may mean for the future of reliable, trustworthy AI.
CV
Christos Diou is an Associate Professor of Artificial Intelligence and Machine Learning at the Department of Informatics and Telematics, Harokopio University of Athens. He is also an affiliated researcher of the Archimedes AI unit and an ELLIS member. He holds a Ph.D. and a Diploma in Electrical and Computer Engineering from the Aristotle University of Thessaloniki. His main research interests include methods for learning robust representations that are capable of out-of-distribution generalization, methods for fair and interpretable machine learning, as well as the use of machine learning for the estimation of causal effects from observational data. He has published several papers in international scientific journals and conferences including venues such as TPAMI, ICLR, CVPR, ICCV, AISTATS and ECAI and has several years of experience participating in and leading European and national research projects, focusing on applications of artificial intelligence in healthcare.
