Parallelization of processes in neural networks

The paper presents the methods of distribution of neurons into the processors of parallel and dispersed processing structures. Neural networks, as executed in parallel, force frequent transfers of messages and data between neurons. When performing the distribution of neurons, the following factors should be taken into account: the optimization of acceleration resulting from parallel conversion and the associated uniform loading of processors. In addition, the number of messages and data transferred between the processors should be considered. The time of inter-processor transmission depends not only on the size of information being transferred, but also on the computational power of contacting converters. The paper does not include computational power analysis, being limited only to the determination of the components of the time of creating and transmitting information.