Xiang Yang, Jensen Finn V, Chen Xiaoyun
University of Guelph, Guelph, ON N1G 2W1, Canada.
IEEE Trans Syst Man Cybern B Cybern. 2006 Jun;36(3):546-58. doi: 10.1109/tsmcb.2005.861862.
This paper extends lazy propagation for inference in single-agent Bayesian networks (BNs) to multiagent lazy inference in multiply sectioned BNs (MSBNs). Two methods are proposed using distinct runtime structures. It was proved that the new methods are exact and efficient when the domain structure is sparse. Both improve space and time complexity more than the existing method, which allows multiagent probabilistic reasoning to be performed in much larger domains given the computational resource. The relative performances of the three methods are compared analytically and experimentally.
本文将单智能体贝叶斯网络（BN）推理中的延迟传播扩展到多智能体多分段贝叶斯网络（MSBN）中的延迟推理。提出了两种使用不同运行时结构的方法。结果证明，当领域结构稀疏时，新方法是精确且高效的。这两种方法在空间和时间复杂度方面都比现有方法有更大改进，这使得在给定计算资源的情况下，能够在大得多的领域中进行多智能体概率推理。通过分析和实验比较了这三种方法的相对性能。
