The synthesis of structural diagrams of automatic devices on formal neurons

Abstract

The development of finite state machines and the synthesis of neural networks come with enormous computational difficulties. The problems that are faced both by the creators of control finite state machines and the creators of neural networks are almost the same. In order for a control finite state machine to be implemented, an algorithm for its operation must be created, and then a program must be written, and finally this program must be implemented in hardware in the form of a finite state machine. It is crucial to create a finite state machine, which will be deterministic. As for neural networks, it is necessary either to set the weights on its edges with the help of experts, or it must be trained to obtain optimal weights on its edges. Both tasks, that is, the determination of finite state machines and the training of neural networks, are currently most often performed using approximate (exponential or genetic) algorithms. At the same time, few authors point out the fact that, firstly these algorithms give an error of up to 15 %, and secondly the operating time is quite long and requires large energy costs. The article has proven that control finite state machines and neural networks are equivalent based on their structure, which can be represented as a directed edge graph. Such equivalence makes it possible to use methods of normalizing arbitrary graphs to determine finite automata and synthesize neural networks. Methods of graph normalizing are extremely new, they are based on a fundamentally new approach of the extension of graph theory and will allow performing these operations using algorithms that have linear complexity or can significantly reduce the number of options when using brute force.

Keywords

• finite machine
• determination
• neural network
• directed graph
• normal algorithm