Work

A New Bounding Scheme for Influence Diagrams

Year: 2021

Type: conference-paper

Abstract: Influence diagrams provide a modeling and inference framework for sequential decision problems, representing the probabilistic knowledge by a Bayesian network and the preferences of an agent by utility functions over the random variables and decision variables. Computing the maximum expected utility (MEU) and the optimizing policy is exponential in the constrained induced width and therefore is notoriously difficult for larger models. In this paper, we develop a new bounding scheme for MEU that applies partitioning based approximations on top of the decomposition scheme called a multi-operator cluster DAG for influence diagrams that is more sensitive to the underlying structure of the model than the classical join-tree decomposition of influence diagrams. Our bounding scheme utilizes a cost-shifting mechanism to tighten the bound further. We demonstrate the effectiveness of the proposed scheme on various hard benchmarks.

Source: Proceedings of the AAAI Conference on Artificial Intelligence

DOI: 10.1609/aaai.v35i13.17443

Authors: Radu Marinescu, Junkyu Lee, Rina Dechter

Institution: University of California, Irvine

Language: English

Topic: Bayesian Modeling and Causal Inference

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: diamond

Funder: National Science Foundation

Award: RI: Small: Anytime Algorithms and Bounds for Probabilistic Graphical Models