Neural Network Algorithm And Application Of Interval-oriented Analysis

Artificial neural network is a kind of information processing system with brain-style.It is based on neuroscience research findings, processing, memory and processing information by simulating the human nervous system.It's features is distributed information storage and parallel co-processing. Traditional neural network'input / output is usually determined number. For some applications, the sample used to train the network could not be accurately given, but only to a certain range given in the form, which gives the traditional neural network makes it difficult. With the range of technology, there has been interval algorithm and neural network combination.Interval algorithm makes stored data as a range, and then operates the interval, so that it is possible to avoid the arithmetic errors from calculation of floating-point. In addition, the interval algorithm makes the interval parameters directly be included in the calculations, it is very important in the practical application. After 20 years of development, the seventies of the last century the interval analysis method can be used to solve nonlinear equations, particularly, it is the effective solution of the error estimates in the presence of the inspection of reconciliation.With the use of interval analysis global optimization of neural network and neural network-based isolation of polynomial real roots, which is a simple combination between neural networks and interval algorithm.This paper analyzed the neural network-based'polynomial real root isolation defects, then improved and used advantage of BFGS variable metric method decreased the advantages of speed in order to quickly find the root of the approximate solution. In this way, we can avoid the whole interval iterative algorithm to improve the computing efficiency of the algorithm.As the graphics processor (GPU) capabilities and rapid development of the programmable rendering pipeline of high-speed and parallelism, making graphics processor general purpose computation (GPGPU) has become a hotspot. It is another work of the paper, using the powerful GPU floating-point computing power and highly parallel computing features to solve this algorithm. Finally, the algorithm with a parallel will further improve the efficiency.