Functional Causal Bayesian Optimization - Lecture 3

Explore a cutting-edge approach to causal inference and optimization in this conference talk from the Uncertainty in Artificial Intelligence (UAI) 2023 conference. Dive into the concept of functional causal Bayesian optimization (fCBO), a novel method for identifying optimal interventions in known causal graphs. Learn how fCBO extends existing Causal Bayesian Optimization techniques to incorporate functional interventions, allowing variables to be set as deterministic functions of other graph variables. Understand the use of Gaussian processes with inputs defined in reproducing kernel Hilbert spaces to model unknown objectives and compute distances between vector-valued functions. Discover the method's ability to sequentially select functions for exploration by maximizing expected improvement while maintaining computational tractability. Examine graphical criteria for determining when functional interventions can lead to improved target effects and conditions for optimal conditional target effects. Gain insights into the method's practical applications through both synthetic and real-world causal graph examples presented in this 24-minute oral session.