No abstract

No abstract

Research and development in modeling and simulation of human cognizance functions requires a highperformance computing platform for manipulating largescale mathematical models. Traditional computing architectures cannot fulfill the attendant needs in terms of arithmetic computation and communication bandwidth. In this work, we propose a novel hybrid computing architecture for the simulation and evaluation of largescale associative neural memory models. The proposed architecture achieves very high computing and communication performances by combining the technologies of hardware-accelerated computing, parallel distributed data operation and the publish/subscribe protocol. Analysis has been done on the computation and data bandwidth demands for implementing a large-scale Brain-State-in-a-Box (BSB) model. Compared to the traditional computing architecture, the proposed architecture can achieve at least IOOX speedup. I. INTRODUCTIONWith the recent ongoing research in human intelligence, more attention has been paid to the autoassociative and heteroassociative neural memory models [2] because in many aspects, their working mechanisms are very similar to the functionality of the cerebral cortex, i.e., neocortex. To evaluate the feasibility and performance of using these models for a complete cognitive function, for example vision, we need to build and simulate a large-scale model that may consist of hundreds of thousands of individual models and a massive number of connections among them. The traditional computing architecture, i.e., a "general-purpose CPU plus centralized memory" cannot fulfill the arithmetic computation and data bandwidth demands to simulate large-scale cortical models. More and more research tends to show that the neocortex follows a hierarchical architecture. At the lowest level of this hierarchy we find the neuron; and moving upward in

No abstract

No abstract