Parallel Implementation of Bayesian Network Structure Learning Algorithms

Timothy O'Neil, University of Akron

Usage Details

Joe Haddad will modify existing Bayesian network learning algorithms for use within a parallel environment. Construction and scoring of Bayesian networks is NP-complete, but heuristic algorithms have been developed to reduce the search space. The K2, Markov Chain Monte Carlo and Bayesian Network Power Constructor learning algorithms will be studied by the intern and then implemented using CUDA. We will then use these algorithms in conjunction with large genomic data sets to infer interactions among genes and groups of genes using Blue Waters. Haddad will document all steps involved with the optimization and utilization of these algorithms and will publish his findings.