Reliable Cause-Effect Quantification in Complex Dynamical Systems for Socially Beneficial Policies

Tackling critical global challenges - climate change, personalized healthcare, or pandemic preparedness - demands robust quantification and understanding of 'if-then' relationships in complex spatio-temporal systems. Traditional dynamical models are limited by oversimplifications and computational constraints, while powerful hybrid models leveraging machine learning (ML) lack the interpretability and generalization capabilities essential for entrusting them with policy guidance. Filling this glaring research gap, DYNAMICAUS develops novel methods for reliable causal inference in such systems, overcoming existing limitations. I will establish a theoretical foundation for causal effects in hybrid dynamical models, develop robust uncertainty quantification techniques under unseen interventions augmented by active data acquisition to reduce uncertainty, and quantify the overall impact of steerable inputs on target outcomes. My methods will be validated across critical domains - climate modeling, treatment planning, and epidemic simulations - demonstrating universal applicability and profound potential for informing policy on pressing societal challenges. By integrating an ethicist directly into the research process from the outset, I proactively address societal implications, ensuring my methodological advancements are positioned to achieve positive social impact. With my unparalleled track record of leveraging machine learning expertise for social good by pushing the boundaries in scientific applications as well as my pioneering work on extending the scope of causal inference and hybrid dynamical systems modeling, this timely project is uniquely positioned to break new ground at the intersection of dynamical systems, causality, and socially beneficial machine learning.