Analysis of Parallel Bayesian Network Learning

摘要

Utilizing Bayesian networks to infer the relationships among genes has proven useful in providing information about how gene interactions influence life. However, Bayesian network learning is inherently slow due to the nature of the algorithm. Search space reductions such as K2 help to speed up this process, but when the need for many Bayesian networks arises, there are limited options for increasing speed. Parallelizing the generation of multiple Bayesian networks across multiple cores leads to linear speed-up with minimal overhead. Spanning the problem across multiple nodes (systems, computers) also brings linear speed-up, but can introduce additional overhead. In this paper, methods are developed to span multiple cores and nodes in order to generate many Bayesian networks in parallel. These methods are then tested to determine the speed increase resulting from their utilization.