Title:  Causal Methods for Understanding Complex Systems

Committee: 

Dr. Davenport, Advisor

Dr. Rozell, Co-Advisor  

Dr. Bloch, Chair

Dr. Gupta

Dr. Varshney

Abstract: The objective of the proposed research is to use causal reasoning to explain complex systems. The work consists of three aims, each seeking to understand a different type of system. In the first aim, I propose a framework for explaining black-box classifiers based on a causal model of how the data and classifier output are generated. The explanation consists of a low-dimensional representation of the data in which a subset of the dimensions represent data aspects that have a large causal effect on the classifier output. In the second aim, I propose new theoretical and algorithmic techniques for inferring the causal structure of coupled dynamical systems. Despite being used to describe the data-generating process of many real-world physical and social processes, most standard methods for causal inference do not apply to these systems. I approach this problem using concepts from causality, embedding theorems, and information theory and develop tools to explain the underlying structure in these systems. In the third aim, I examine the question of when and how humans are biased consumers of information by developing new causal models that mathematicize hypotheses in social science literature. The proposed work in this aim could suggest better methods for communicating policy-relevant scientific knowledge to the public. Taken together, the proposed work demonstrates the value of causal reasoning in understanding computational, physical, and social systems.