Research On Extreme Learning Machine And Its Application Based On Manifold Learning

With the development of technology and computer industry,some algorithms for classifying data samples have emerged one after another in recent years.Extreme learning machine(ELM)stands out from these classification algorithms because of its simple structure and simple calculation.As an efficient learning algorithms,it is extended on the basis of single hidden layer neural network.ELM algorithm has indeed achieved satisfactory results in the process of application but there are still many limitations.In order to better improve the generalization ability of ELM and the accuracy of experimental results and solve the problem of insufficient learning due to insufficient data samples,this paper will combine extreme learning machine and manifold learning algorithms to propose two different algorithms.The main research is as follows:(1)We propose a supervised dimensionality reduction algorithm(SNPE)based on the NPE algorithm and the ELM algorithm.In the SNPE algorithm,we introduce the inter-class dispersion matrix with discriminant information into the NPE model.The objective function is minimized to achieve the purpose that minimize the distance of the same category sample points and maximize the distance of different category sample points,ELM algorithm is used to classify data samples to avoid overlap of sample points,which obtains satisfactory experimental results.While retaining the advantages of NPE algorithm,SNPE algorithm fully mines the hidden discriminant information,thus improving the generalization performance of the algorithm.(2)We propose an extreme learning machine based on popular learning(NPELM),which combines the ELM algorithm and the intra-class and inter-class dispersion matrix.The information difference matrix(with the difference between the inter-class dispersion matrix and the intra-class dispersion matrix)is added to the objective function of the ELM algorithm.Which essentially optimize the output weight of the ELM algorithm to improve the ELM algorithm generalization Performance and classification capabilities.Through a series of face database experiments,it is verified that the NPE algorithm has better classification effect than other algorithms,then further proves that the improved algorithm is effective and feasible.