INNE: a Neural Network Simulation Environment

Maria Alberta Alberti - Ivan Serina

Abstract

This paper presents the Interactive Neural Network Environment INNE, a graphical environment to design, simulate and analyse the behaviour of neural networks. The aim of the application is to provide a rich and flexible tool for learning neural network models. Therefore the environment provides several different neural models and tools for visualizing the learning phase and the results of the computation. This last feature makes INNE rather different from other simulators and makes it suitable for students at their first approach to neural networks or to a specific model. Nevertheless the environment is general and robust and allows the design of neural networks to solve problems and to verify their performances. At present the models implemented are: Boltzmann machines and Hopfield networks, Back-error propagation networks, Hebbian networks, Simple competitive networks, Kohonen networks. The application has been developed in the CoLoS project.

 
• Introduction
  • INNE architecture
• Simulation of Boltzmann machines
• Simulation of Back-error propagation Algorithms
• Simulation of Hebbian Algorithms
  • The example of image compression
• Simulation of competitive networks
  • The example of colour clustering
• Simulation of Kohonen's maps
• Acknowledgments
• References