Parallel Sampling for Efficient High-dimensional Bayesian Network Structure Learning
This is an incremental improvement for researchers and practitioners dealing with high-dimensional Bayesian network learning, offering better performance within computational constraints.
The paper tackles the computational expense of approximate Bayesian network structure learning in high-dimensional data by proposing PS-MINOBS, an algorithm that performs parallel sampling on candidate parent sets, and it discovers higher score structures than MINOBS in most cases under the same runtime limit.
Score-based algorithms that learn the structure of Bayesian networks can be used for both exact and approximate solutions. While approximate learning scales better with the number of variables, it can be computationally expensive in the presence of high dimensional data. This paper describes an approximate algorithm that performs parallel sampling on Candidate Parent Sets (CPSs), and can be viewed as an extension of MINOBS which is a state-of-the-art algorithm for structure learning from high dimensional data. The modified algorithm, which we call Parallel Sampling MINOBS (PS-MINOBS), constructs the graph by sampling CPSs for each variable. Sampling is performed in parallel under the assumption the distribution of CPSs is half-normal when ordered by Bayesian score for each variable. Sampling from a half-normal distribution ensures that the CPSs sampled are likely to be those which produce the higher scores. Empirical results show that, in most cases, the proposed algorithm discovers higher score structures than MINOBS when both algorithms are restricted to the same runtime limit.