Node Level Primitives for Exact Inference using GPGPU17th International Conference on Systems, Signals and Image Processing (2010)
AbstractExact inference is a key problem in exploring probabilistic graphical models in a variety of multimedia applications. In performing exact inference, a series of computations known as node level primitives are performed between the potential tables in cliques and separators of a given junction tree. The computation complexity increases dramatically with the clique width and the number of states of random variables. In this paper, we propose a conflict-free data layout for potential tables on GPU. We map the algorithms for the primitives to the GPU architecture based on the proposed data layout. Several optimization techniques are presented to improve the performance. We implemented the proposed method on NVIDIA Tesla C870. Experimental results exhibit scalability over a wide range and show superior performance compared with state-of-the-art multicore CPUs such as Intel Xeon and AMD Opteron.
Citation InformationHyeran Jeon, Yinglong Xia and Viktor K. Prasanna. "Node Level Primitives for Exact Inference using GPGPU" 17th International Conference on Systems, Signals and Image Processing (2010)
Available at: http://works.bepress.com/hyeran_jeon/6/