A neural approach for solving the constraint satisfaction problem

The realized work consists in modeling and solving a constraint satisfaction problem (CSP) by a neural approach. We intend to develop an algorithm based on the conception of a basic neural network able to solve some instantiations of the CSP. The variables are associated to the input and the output nodes of the network and the constraints correspond to the nodes of the hidden layers. The obtained results show that the network can be trapped in local minima. Therefore, we intend to modify the way of calculating the weights of the input layer, so as to improve the structure of the network initially conceived.