An efficient approach for multiple probabilistic inferences with Deepwalk based Bayesian network embedding

Abstract

As a classical probabilistic graphic model, Bayesian network (BN) is widely used for representing and inferring dependence relationships with uncertainties. However, multiple probabilistic inferences on BN are quite inefficient, since each probabilistic inference on BN is extremely time-consuming and meanwhile the intermediate results cannot be reused. It is necessary to improve the overall efficiency of multiple probabilistic inferences on the same BN by incorporating an easy-to-calculate inference method and an easy-to-reuse technique for common calculations in multiple inference tasks. In this paper, we first propose a Deepwalk based method for BN embedding, as a specification of general graph embedding, which preserves both the graphical structure and conditional probabilities in BN. We then provide the algorithm to approximate probabilistic inferences via the distance among embeddings. We finally present an efficient approach to multiple probabilistic inferences. Extensive experiments illustrate that our proposed method is effective for BN embedding, and outperforms other state-of-the-art competitors by improving the inference efficiency with several orders of magnitude.