Parallel Implementation of Logical Neural Networks Using CUDA Technology

In this work we consider the logical neural networks and their method of implementation in the programming language C ++ with CUDA technology for parallel execution of operations within the network. Purpose – an implementation of a neural network, as a system of logical inference. In the process of work was considered resolution method of Robinson, which is used as a base for many inference systems, and modification of this method for parallel computations made by Ehud Shapiro. In this work was implemented a parallel algorithm of logical implication in the language C ++, allows operation of logical implication simultaneously on the entire set of disjuncts. To implement the chosen algorithm parallelization CUDA technology was used, because of allowing the use of a graphics card as an additional processor. Due to the architecture of the graphics card, which contains a processor ten times more cores than the CPU, it is possible to use them for any simultaneous parallel operations. Use of a graphics card for implementing the algorithm allows not only achieving performance at the expense of parallelism, but freeing from the CPU load, allowing other processes may be executed on the CPU. With the implementation of the parallel algorithm has been found a reduction of the time of work of the program because of replacing of sequential selection of disjuncts to a parallel manner. As a result of the work it was implemented logic neural network, which uses the graphics card processor for parallel logical implication. Due to the good scalability implemented logic neural network it can be used not only to solve the problem of logical implication, but is modified in a variety of control systems.