*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Title: On the Recovery of Causal Graphical Models from Observational and Interventional Data
Committee:
Dr. Fekri, Advisor
Dr. Bloch, Chair
Dr. Mao
Abstract: The objective of the proposed research is to develop novel algorithms for learning causal relationships from both observational and interventional data using techniques from variational inference, normalizing flows and reinforcement learning. In addition, the focus is also on developing the underlying theory for proving identifiability under certain assumptions as well as proving consistency and correctness of the proposed methods. Learning causal relationships between variables is a fundamental problem in science as causal understanding allows for predicting the behavior of systems under unseen conditions. However, this remains a challenge as causal learning is an NP-Hard problem, and it is in general not identifiable. Although identifiability results exist under certain assumptions, these assumptions can be severely limiting when applying these methods on real-world datasets. The goal of this research is to contribute to the existing literature on causal discovery and develop causal learning algorithms that work well in practice. To that end, preliminary results are demonstrated for the following two settings: (1) causal graph recovery from compressed measurements, and (2) nonlinear cyclic causal graph learning, which achieve improved performance over the baselines in the conducted experiments. Further details on the future direction of research are then provided for rendering causal discovery methods practical, while being theoretically grounded.
