Research On Probability Graph Model Reasoning Algorithm And Parallelization

Because probabilistic graph inference and learning have been widely applied in the latest research fields of computer vision,Natural Language Processing,speech recognition and bioinformatics,it has become a research hotspot in academia and industry.How to improve the speed of probability graph model reasoning algorithm has become an important research direction in the field of probability graph model.In recent years,the general computing technology of the GPU has been developed rapidly.GPU has become the mainstream computing platform in the field of high performance computing.Based on the above situation,the existing parallel acceleration algorithm is fully studied.After referring to the idea of reasoning algorithm based on matrix computing,CUDA computing framework has been used to achieve parallel acceleration of GPU based probabilistic graph model reasoning algorithm.This paper mainly deals with the parallelization of the exact reasoning algorithm in the probability graph model reasoning algorithm.Firstly,the probability graph model is studied in depth,and the precise reasoning algorithm is focused on,and the parallel computing steps can be found in these algorithms.Then,in order to adapt to the parallel computing framework of CUDA GPU,we analyze the related algorithms of exact reasoning,and a probabilistic model reasoning algorithm based on GPU is proposed.Finally,we test the probabilistic graph model reasoning algorithm based on GPU,node level parallelization and topology parallelization,and compare it with the serial algorithm on CPU,and analyze the experimental results,and illustrate the effectiveness of the algorithm.