Data Parallelism for Belief Propagation in Factor Graphs

Nam Ma, Yinglong Xia, Viktor K. Prasanna
2011 2011 23rd International Symposium on Computer Architecture and High Performance Computing  
We investigate data parallelism for belief propagation in acyclic factor graphs on multicore/manycore processors. Belief propagation is a key problem in exploring factor graphs, a probabilistic graphical model that has found applications in many domains. In this paper, we identify basic computations called node level primitives for propagating the belief in a factor graph. Algorithms for these primitives are developed using data parallel techniques. We propose a complete belief propagation
more » ... ithm using the primitives to perform exact inference in factor graphs. We implement the proposed algorithms on state-of-the-art multi-socket multi-core systems with additional NUMA-aware optimizations. Our proposed algorithms exhibit good scalability using a representative set of factor graphs. On a four-socket Intel Westmere-EX based system with 40 cores, we achieve 39.5× speedup for the primitives and 39.2× for the complete algorithm using factor graphs with large potential tables.
doi:10.1109/sbac-pad.2011.34 dblp:conf/sbac-pad/MaXP11 fatcat:ziclr2joovhn7jeczbpbyomumi