糖心TV

Abstract:
Understanding cause-and-effect relationships is a fundamental challenge across the data sciences. Traditional measures, such as the Average Causal Effect (ACE), summarise causal relationships through averages but often obscure important heterogeneity across individuals or subpopulations. In this talk, we introduce new causal effect measures that capture and interpret this heterogeneity to support causal explanation and decision-making. We characterise causal heterogeneity through statistical moments—variance, skewness, and kurtosis—present new formulations for the probabilities of causation, and introduce new metrics for causal decision-making under multiple actions. We outline identification and bounding results under common causal assumptions and illustrate the application of these measures in real-world settings.

Bio:
Jin Tian is a Professor of Machine Learning at MBZUAI and Editor-in-Chief of the Journal of Causal Inference. His research focuses on causal modelling, probabilistic graphical models, and adversarial robustness, including work on probabilities of causation and causal effect identification in cluster DAGs. He is also an Action Editor for the Journal of Machine Learning Research and previously spent nearly two decades on the Computer Science faculty at Iowa State University. He earned his PhD in Computer Science from UCLA after completing a bachelor鈥檚 degree in Physics at Tsinghua University.