Data Parallel Implementation of Belief Propagation in Factor Graphs on Multi-core Platforms

摘要: We investigate data parallel techniques for belief propagation in acyclic factor graphs on multi-core systems. Belief is a key inference algorithm graph, probabilistic graphical model that has found applications many domains. In this paper, we explore parallelism basic operations over the potential tables propagation. Data these table are developed shared memory platforms. then propose complete using to perform exact graphs. The proposed algorithms implemented state-of-the-art multi-socket systems with additional NUMA-aware optimizations. Our exhibit good scalability representative set of On four-socket Intel Westmere-EX system 40 cores, achieve 39.5 $$\times $$ × speedup and 39 large tables.