Research And Application Of Manifold Learning And Classification Algorithm

Manifold Learning(ML)is one of the methods of pattern recognition,which has broad development prospects and application value in image processing,artificial intelligence,natural language processing,etc.The feature of most real high-dimensional data points generally is nonlinear in machine learning.We can use the manifold learning method to reduce the dimension and extract local features to achieve data visualization or data reduction.The main idea of Manifold Learning is to project the high-dimensional spatial sample set into the low-dimensional manifold subspace,which preserve the local neighborhood manifold structure of the original sample set and find the rule within the data.Dimensionality reduction is the most important technical manifestation of manifold thought.Researchers have made some research results on the dimensionality reduction algorithm based on Manifold Learning,such as Laplacian Eigenmap(LE),Locally Linear Embedding(LLE),Isometric mapping(Isomap)and Neighboring Preserving Embedding(NPE),etc.In pattern recognition technology,the dimensionality reduction algorithm based on Manifold Learning has been widely used in face recognition and speech recognition.Face image data and speech data have features of high spatial dimension,sparsity,and uneven distribution.To improve the recognition performance of face data,based on manifold learning principle and technique,we have researched some related dimensionality reduction technology based on manifold learning and proposed a new neighborhood preservation embedding algorithm(RNPE).The algorithm emphasizes the discriminative information within the data,and makes the data after dimension reduction achieve optimal separation by maximizing the inter-class dispersion and minimizing the intra-class dispersion,which effectively mines,extracts and retains the local manifold topological structure of the data.In addition,we have studied the Extreme Learning Machine(ELM),and used it to classify with different algorithms.The experimental result shows that the proposed algorithm has higher recognition accuracy than the traditional algorithms.