The goal of this project is to develop supervised learning algorithms for feed forward and RBF neural networks, and a novel method for data clustering to perform classification based on a set of potential fields synthesized over the domain on input space by a number of potential function units. A fundamental component in building neural network architectures with potential functions is the determination of potential function entity (PFE), which is designed to generate a respective potential function over the domain of input space. It is assumed that a certain input vector of the input space is comparable to an electrical charge, located at the same position, which generates an electrostatic potential. In this research, all inputs carry charges and generate potentials to all other inputs.

We focus on the design of:

• The basic topologies of the feed forward and RBF neural networks and building criteria for determination of the number of hidden layers and number of hidden units. These neural network architectures should incorporate structural changes during training which will further improve the convergence rate of the neural networks.
  
  
• Learning algorithms for both supervised neural networks with dynamic allocation of the number of hidden neurons during the learning process. Clusters of the same object type should be established in such a way that they should generate a higher accumulated potential than all other classes. When the learning procedure is completed, the neurons of the neural networks should produce strong responses (high potential) to those training objects, which belong to the same class, and produce weak responses to the training objects of the other classes.