Genetic algorithm finds simple consensus Bayesian networks with limited complexity
Structural Fusion of Bayesian Networks with Limited Treewidth Using Genetic Algorithms
Neural and Evolutionary ComputingMachine Learning
Summary
Combining many Bayesian networks into one can be hard because the combined network might become too complex to use for quick decisions. The authors introduce a genetic algorithm, which is a way of trying many solutions and keeping the best ones, to create a simpler combined network that still captures important information from the originals. They keep the network complexity low by limiting a property called treewidth, which helps make sure computations stay manageable. Their experiments show this method can effectively merge networks while keeping them practical.
What this means in practice
- •For data integration teams: Merge multiple complex Bayesian models into a single simpler model without losing key dependencies, enabling faster probabilistic reasoning.
- •For risk management analysts: Create combined risk assessment models from different sources ensuring the resulting model runs efficiently on standard software.
Authors
Pablo Torrijos, José A. Gámez, José M. Puerta
Abstract
This paper introduces an evolutionary computation approach for consensus in structural Bayesian Network (BN) fusion under the constraint of limited treewidth. The consensus BN aims to reconcile multiple input BNs into a single one that retains key structural features present in the original networks. Treewidth, a graph-based parameter associated with computationally tractable inference, is utilized to restrict the complexity of the resulting network. A genetic algorithm is proposed to look for a BN that codifies as much information about the unrestricted fusion as possible while ensuring the treewidth restriction. Experimental evaluation demonstrates the genetic algorithm's ability to obtain consensus BNs with limited treewidth, providing a valuable tool for aggregating information from diverse sources while returning a computationally actionable model.