Research On Extreme Learning Machine Algorithm Based On Dimensionality Reduction

With the development of computer and multimedia information technology,the field of machine learning techniques has also developed rapidly in recent years.As one of the research hotspots of machine learning,the extreme learning machine(ELM)attracts the attention of many researchers with its simple theory and easy implementation.ELM has high efficiency and good generalization ability,and has been widely used in classification and regression issues.However,there are still many shortcomings in dealing with specific issues.For example,in the face of high dimensionality,noise,and outlier data,the classification accuracy of ELM is reduced.In the case of a limited number of data samples,for supervised learning,ELM will have the problem of insufficient learning.In this paper,we give a research on ELM aiming to deal with the concerned problems.The main research results are as follows:(1)In order to solve the influence noise and redundant attributes of face image on the ELM algorithm,we proposes an ELM based on manifold learning.We introduce the discriminative information of the data into the locality preserving projections(LPP),achieving a locality preserving discriminant projections(LPDP).LPDP introduces intra-class discriminant information and inter-class discriminant information of data into the LPP model,and optimizes the LPP model.LPDP not only inherits the advantages of LPP,but also takes into account the discriminating information of the data and obtains better dimensionality reduction effects,thereby improving the generalization capability of ELM.(2)Aiming at to deal with the issue that the existing ELM and its improved algorithms can not make good use of the discriminative information contained in the data,we proposes a discriminative information regularized extreme learning machine(IELM)based on discriminative information.IELM introduces the concept of similar dispersion and heterogeneous dispersion,which is reflected in the discriminating information of input spatial data.By maximizing the dispersion of heterogeneous clusters and minimizing the dispersion of similar clusters,the output weights of ELM are optimized,which improves the classification performance and generalization ability of ELM to some extent.Experimental results show that the two optimized algorithms improve the classification effect of ELM significantly,and have better classification accuracy and generalization ability than other algorithms.