Shuyang Bai (University of Georgia)

Extremal structural causal model, causal Markov property and causal direction identification

Monday 14 September at 13:00 UTC time = (09:00 New York) = (15:00 Paris) = (21:00 Shanghai)

Zoom Link 

In this talk, we discuss a formulation of structural causal models tailored to multivariate extremes. In contrast to conventional structural causal models, in which randomness is governed by a probability law, our structural equations are governed by an exponent measure, an infinite measure on a punctured space that characterizes the dependence structure of extreme observations. We show that this formulation is compatible with a recently introduced notion of conditional independence defined in terms of the exponent measure. We also discuss natural assumptions arising from this framework that enable the identification of causal directions. 

The talk is based on joint work with Fei Fang, Vishal Routh, and Tiandong Wang.