Talk by Asst. Prof. Arnab Bhattacharyya from National University of Singapore: "Learning and Testing Causal Models with Interventions"
Learning and Testing Causal Models with Interventions
Asst. Prof. Arnab Bhattacharyya
National University of Singapore
10:30am / December 5th, Thursday
15F meeting room #1512, NII
Causality has long been a subject of study in philosophy, logic,
computer science and beyond. We consider the framework of causal
Bayesian networks as defined by Pearl (2009). In this setting, we obtain
efficient algorithms with bounded number of samples for goodness-of-fit
testing, two-sample testing, and learning discrete causal Bayesian
networks on given graphs of bounded degree and bounded "confounded
components". Our algorithms sample from a small number of interventions
of the input causal models, in contrast to previous work which don't
establish any finite sample bounds.
Joint work with Jayadev Acharya, Constantinos Daskalakis and