Parallel cycle-based branch-and-bound method for Bayesian network learning

Abstract

Bayesian networks (BNs) are one of the most commonly used models for representing uncertainty in medical diagnosis. Learning the exact structure of a BN is a challenging problem. This paper proposes a multi-threaded branch-and-bound (BB on the other hand, the learning procedure is alleviated by executing CB-B&B over a set of parallel processors. In comparison with conventional exact structure learning approaches for BN, the obtained results demonstrate that the proposed CB-B&B is efficient. On average, it produces the exact structure for BN three times faster than the standard B&B version. We also present simulations on parallel CB-B&B which show a significant gain in terms of execution time.