Higher Order Bayesian Networks, Exactly - POPL 2024

Explore a 16-minute conference talk from POPL 2024 introducing a higher-order programming language for Bayesian networks. Discover how this lambda-calculus-based language proves sound and complete with respect to Bayesian networks, allowing for the encoding of networks as terms and the compilation of programs into networks. Learn about the language's ability to specify recursive probability models and hierarchical structures, as well as its compositional and cost-aware semantics based on factors. Delve into the advanced techniques from linear logic, intersection types, rewriting theory, and Girard's geometry of interaction that underpin this novel approach to probabilistic programming and Bayesian inference.