Decision Diagrams for Efficient Inference and Optimization in Expressive Discrete-Continuous Domains

Explore an in-depth lecture on the extension of algebraic decision diagrams (ADDs) to continuous variables, introducing the extended ADD (XADD) for representing piecewise functions. Learn about efficient computation methods for elementary arithmetic operations, integrals, and maximization within various function restrictions. Discover the wide-ranging applications of XADDs, including novel closed-form solutions in probabilistic inference for hybrid graphical models, parametric constrained optimization, sequential decision-making in continuous state and action domains, and joint prediction and optimization in machine learning. Gain valuable insights from Scott Sanner of the University of Toronto in this 45-minute talk, part of the Probabilistic Circuits and Logic series at the Simons Institute.

Decision Diagrams for Efficient Inference and Optimization in Expressive Discrete+Continuous Domains