Multicore cluster implementations of hierarchical Bayesian cortical models

We examine the parallelization of two recent biologically inspired hierarchical Bayesian cortical models onto two multicore processor based clusters. The models examined have been developed recently based on new insights from neuroscience and have several advantages over traditional neural network models. In particular, they need far fewer network nodes to simulate a biological scale cortical system than traditional neural network models, thus making them computationally more efficient. The two architectures examined are the Sony/Toshiba/IBM Cell BE and the Intel quad-core Xeon processors. Our results indicate that optimized implementations of the models on clusters of multicore processors can provide significant speedups and that such clusters are a promising approach for developing large scale simulations of the models. We show that for small scale implementations of the models, multicore clusters can provide speedups of about 850 times over serial implementations on the Cell Power Processor Unit.