Shortest Path Based Decision Making Using Probabilistic Inference
DOI:
https://doi.org/10.1609/aaai.v30i1.9909Abstract
We present a new perspective on the classical shortest path routing (SPR) problem in graphs. We show that the SPR problem can be recast to that of probabilistic inference in a mixture of simple Bayesian networks. Maximizing the likelihood in this mixture becomes equivalent to solving the SPR problem. We develop the well known Expectation-Maximization (EM) algorithm for the SPR problem that maximizes the likelihood, and show that it does not get stuck in a locally optimal solution. Using the same probabilistic framework, we then address an NP-Hard network design problem where the goal is to repair a network of roads post some disaster within a fixed budget such that the connectivity between a set of nodes is optimized. We show that our likelihood maximization approach using the EM algorithm scales well for this problem taking the form of message-passing among nodes of the graph, and provides significantly better quality solutions than a standard mixed-integer programming solver.