Distributed launch of neural networks on a set of computing nodes

Abstract

The task to organize neural network on a set of computing nodes in a cluster environment is defined. A solution is proposed to get a result close to the optimum. The analysis of the extreme cases, when the algorithm deviates from the optimum result, is made. The application of the algorithm is presented in a concrete implementation that runs in a clustered environment. The issues and assumptions that should be considered in further development of the proposed algorithm are described.