A Constrained Learning Algorithm for Finding Multiple Real Roots of Polynomial

A constrained learning algorithm is proposed for finding the multiple real roots by neural networks based on the complete discrimination system of polynomial. By coupling a constrained condition into the error function of the neural networks, the proposed algorithm is effectively to avoid the weights fluctuating in a large range. To speed up the convergence speed, a momentum term is added into the learning algorithm. Experiment results show that the presented constrained algorithm is of not only faster convergent speed, but also more effectiveness comparing to the unconstrained one.