Approximate Inference Algorithm Based on Adaptive Ant Colony and Artificial Fish Swarm Algorithm for Credal Networks

Abstract

This paper presents a new algorithm for approximate inference in credal networks (that is, models based on directed acyclic graphs and interval-valued probabilities). Approximate inference in credal networks can be considered as multistage decision in this paper. It is looked as combinatorial optimization problems that obtaining the extreme posteriors from the combinations of various vertices in credal networks. Based on this, the paper combines two intelligence swarm algorithms (ant colony algorithm and artificial fish swarm algorithm) to obtain interval posterior probabilities of query variable for the states of given evidence variables.