Implementation of Neural Networks with Help of a Data Flow Virtual Machine

摘要

The main goal of this paper is to show how a neural network can be implemented with help of the data flow management system at a virtual machine. As an example, the three-layer neural network realization has been investigated to solve a simple XOR function with two inputs and one output. For that purpose, a sigmoid command required to make a neuron activation function has been added into the data flow virtual machine. It is presented in the paper that neural networks can be described as data flows with help of the declarative approach on a base of the JSON format.