Genetic algorithm optimize neural network based on structural risk minimization

The paper demonstrates a method for optimizing a neural network based on structural risk minimization (SRM). The method combines the principle of SRM and neural network using genetic algorithms to optimize the perceptron to avoid the failings of the traditional perceptron; local convergence of connection weight and high probability of failed recognition. Owing to global optimization by the genetic algorithm, the improved method can evolve and adapt itself. It has better generalization performance and better property of avoiding disturbance, which improves the whole performance of the neural network. Compared with the normal algorithm of SVM, it has wide application and strong ability to deal with large data.