EECS Seminar: Causal Graph Inference - New methods for Application-driven Graph Identification, Interventions and Reward Optimization

Abstract:  Causal inference enables understanding of the underlying mechanisms in complex systems, with applications spanning social sciences, economics, biology and machine learning. Uncovering the underlying cause-and-effect relationships facilitates the prediction of the effect of interventions and the design of effective policies, thus enhancing the understanding of the overall system behavior. Causal structures are often represented by Bayesian networks in the form of directed acyclic graphs (DAGs). Herein we explore several problems within causal inference motivated by applications in microbial communities and epidemiology. In particular, these application spaces are governed by a modest number of observations and unconventional underlying observation distributions. For example, graph identification is done via the collection of observations or realizations of the random variables, which are the nodes in the graph. A host of strategies have been proposed for causal graph identification from greedy methods to those based on sparse approximation. Another challenge with many provably correct graph finding methods is their attendant computational complexity. We design graph finding algorithms that control error rates such as the false alarm rate for edge detection as well as low-complexity methods that learn sub-graphs versus the entire graph simultaneously. For both approaches, finite-sample performance analyses are conducted. We then explore the impact on graph identification through the design of interventions:  changing the nature of the observations of nodes in the graph in a controlled, but not always known fashion. We see that interventions can strongly improve graph learning.  Finally, we investigate the causal bandit problem, with the objective of maximizing the long-term reward by selecting an optimal sequence of interventions on nodes in an unknown causal graph. It is assumed that both the causal topology and the distribution of interventions are unknown. Interestingly, reward optimization is sensitive to false negatives in the estimated causal graph. Examples from gene expression in bacteria and zoonotic virus tracking epidemiology are provided.

Bio:  Urbashi Mitra received bachelor's and master's degrees from the University of California at Berkeley and her doctorate from Princeton University. Mitra is currently the Gordon S. Marshall Professor in Engineering at the University of Southern California with appointments in electrical engineering and computer science. Mitra is a fellow of the IEEE, a foreign member of the Academia Europaea and a member of Phi Kappa Phi. She was the inaugural editor-in-chief for the IEEE Transactions on Molecular, Biological and Multi-scale Communications and has held multiple associate editorships for IEEE transactions. Mitra is the recipient of: the 2025 Princeton ECE Department Distinguished Graduate Alumni Award, the 2024 IEEE Information Theory Society Aaron D. Wyner Distinguished Service Award, the 2021 USC Viterbi School of Engineering Senior Research Award, the 2017 IEEE Communications Society Women in Communications Engineering Technical Achievement Award, a 2016 UK Royal Academy of Engineering Distinguished Visiting Professorship, a 2016 US Fulbright Scholar Award, a 2016-2017 UK Leverhulme Trust Visiting Professorship, 2012 Globecom Signal Processing for Communications Symposium Best Paper Award, 2012 U.S. National Academy of Engineering Lillian Gilbreth Lectureship,  the 2009 DCOSS Applications & Systems Best Paper Award, 2002 Texas Instruments Visiting Professor, 2001 Okawa Foundation Award, 2000 OSU College of Engineering Lumley Award for Research, and a 1996 National Science Foundation CAREER Award. Her research interests are in model-based machine learning, wireless communications, communication and sensor networks, biological communication systems, detection and estimation and the interface of communication, sensing and control.