A Probabilistic Programming Language for Influence Diagrams
Refereed Conference Meeting Proceeding
Probabilistic Programming (PP) extends the expressiveness and scalability of Bayesian networks via programmability. Influence Diagrams (IDs) extend Bayesian Networks with decision variables and utility functions, allowing them to model sequential decision problems. Limited-Memory IDs (LIMIDs) further allow some earlier events to be ignored or forgotten. We propose a generalisation of PP and LIMIDs called IDLP, implemented in Logic Programming and with a solver based on Reinforcement Learning and sampling. We show that IDLP can model and solve LIMIDs, and perform PP tasks including inference, finding most probable explanations, and maximum likelihood estimation.
Scalable Uncertainty Management 2017 - 11th International Conference
Lecture Notes in Computer Science 10564, Springer 2017
Digital Object Identifer (DOI):
National University of Ireland, Cork (UCC)
Open access repository: