因果图中高效并行GIBBS仿真算法的研究

基于Markov Chain Monte Carlo(MCMC),思想的Gibbs仿真算法的引入使得大型因果图模型的推理速度得到极大提高，而利用节点间相互独立的特性，可以对其进行并行的采样，从而进一步加快推理速度。该文通过分析Gibbs算法，提出了将整个推理运算过程映射到多处理机系统中的判定准则，防止了机械地对处理机进行分配而造成的计算资源的浪费，算法能够根据实际处理机的数目以及不同的计算能力而灵活地分配计算资源，更加有利于发挥并行机的计算能力。通过仿真实验，验证了计算模型的有效性。

The Research of Parallel Gibbs Simulating Algorithm in Causality Diagram

The introduction of Gibbs simulating algorithm, which is based on Markov Chain Monte Carlo(MCMC) theorem, greatly improved the reasoning speed of causality diagram methodology. However, there is a way to speed up the simulation of variables with Markov structure, that is to exploit the neighbor structure to enable updates of several components independently. After analyzing the Gibbs algorithm in Causality Diagram, this paper put forward a rule which regulates the process of mapping from reasoning calculation to multiprocessor system, and avoids the waste of calculating resources due to the hidebound processor allotting. This algorithm will flexibly allot the calculating resources according to the processor number and different calculating capability, so the parallel calculating performance could be improved. The validity of this algorithm has been proved through a simulating experiment.