Title: Causal Data Science: Estimating Identifiable Causal Effects
Abstract: Causal inference is essential in many disciplines, including statistics, artificial intelligence, and public health. Two foundational tasks in causal inference are (1) causal effect identification, which determines whether causal effects can be computed from available data, and (2) causal effect estimation, which computes these effects from finite samples. While the theory of causal effect identification is well established, effective and robust estimation has been addressed for only a subset of identification scenarios. Fully connecting identification with estimation makes causal inference far more practical.
In this talk, I will present an estimation framework that bridges causal effect identification and estimation. I will begin by introducing a unified estimation framework for any identifiable causal effect from observational data, demonstrating its application in public health and explainable AI. Building on this foundation, I will describe an extended estimation framework that fuses observational and experimental datasets, thereby enabling robust estimation in more complex settings. I will further present a scalable and unified causal effect estimation method that encompasses a range of causal inference scenarios. Finally, I will introduce my lab’s future research pillars toward advancing causal data science.
Short Bio: Yonghan Jung is an Assistant Professor in the School of Information Sciences at the UIUC and the leader of the Causal Data Science Lab. His research focuses on estimating causal effects in complex, imperfect, and large-scale data, with broad applications in trustworthy AI and healthcare science. More information can be found athttps://www.yonghanjung.me/.