Using finite state automats with the aim of improvement neural nets characteristics


© 2007, E.V. Pervushin

Saint-Petersburg State University of Information Technologies, Mechanics and Optics

Project documentation
Application 1
Application 2
Source code of application 1
Source code of application 2

Annotation

The purpose of work is researching an ability of implementation some of the neural net elements in the form of finite state automats. This variant of combining finite state automats and neural net uses automats states as the entrance of neural net, and automats perform prior analyzing of entry data. The aim of this structure is improvement of neural net characteristics, based on finite state automats ability for keeping historical information for entry data.

The proposed method of combining neural net and finite state automats tested on examples below:

  • the recognition system for moving images. At the image marked moved element, its outline and its speed and direction;
  • motion car simulator. The control system of the car analyzes road situation and operates the car to avoid collisions.