The Algorithm Complexity Research Of Second Class Othognal (Tensor) Weight Function Neural Network And Its Application

Weight function neural network is a new kind of neural network developed in recent years,which has many advantages, such as finding globe minima directly, good performance ofgeneralization, extracting some useful information inherent in the problems and so on. Timecomplexity is an important measure of algorithm. Algorithm with a lower time complexity alwaysbrings higher operating efficiency. So study the complexity of the second class orthogonal weightfunction algorithm is very meaningful.This paper is based on the weight function neural network theory and combined with theknowledge of numerical analysis, orthogonal function and algorithm complexity to analyze thecomplexity of second class orthogonal weight function neural network. This paper ultimatelyconcludes that the algorithm time complexity of the second class orthogonal weight function neuralnetwork has relationship with the input dimensions, output dimensions and the sample number. Onthe basis of theoretical analysis, use MATLAB tools to simulate the algorithm complexity of thesecond class orthogonal weight function neural network and proves the correctness of thetheoretical results.There may have samples that with the same input value but different output value whiletraining. Those samples are called singular samples. Singular samples will lead to weight functionunsolvable, resulting in the termination of the neural network training. In order to solve the problem,under the guidance of supervisor this paper preliminary study the singular sample solution based ontensor space transformation which through the axis rotaion to ensure the original singular samplesare non-sinular in the new coordinate system and without new singular samples.On the basis of algorithm complexity analysis this paper applies the second class orthogonalweight function neural network algorithm in speech recogoniton. Speech characteristic signalidentification is an important aspect of speech recognition. The paper applys second classorthogonal function neural network in recognizing four kind of music characteristic signal,preliminary results indicate that second class orthogonal function neural network has a lowcomplexity and good recognition rate in speech recognition.