Extreme Learning Machine Alorithm And Application Research

Machine learning is the core research direction of today's big data Era.The research results of machine learning are widely used in pattern recognition,computer vision,data mining,cybernetics and other fields.Classifier research plays an important role in the research of machine learning.Most of the practical problems can be converted into a classification problem.Because it is very difficult to obtain the required information directly from the original data.The ELM comes from the neural network of hidden layer.With easy realization,fast speed,strong generalization performance.This article through to the ELM thorough research,aiming at the defects in theory and application,different optimization improvement methods are put forward.By this way the performance of the ELM was improved,and the application field was widened.This thesis focuses on summary the ELM's parameter optimization method and development trend of online sequential extreme learning machine(OS-ELM)algorithm.Provided the development of history and the latest results of research directions for the researcher.In order to improve the ELM classification accuracy and generalization ability,the parameters which based on modified differential evolution(MDE)algorithm is proposed.Two main factors that affect the performance of OS-ELM are computational complexity and sample selection.A new OS-ELM based on hybrid kernel function is proposed.The concept of mixed kernel function and the sample selection method based on membership function are introduced into the OS-ELM model.The experiment results demonstrate that the ELM based on MDE has better approximation performance and good generalization performance.Meanwhile,the MDE algorithm to optimize the parameters of hidden layer ELM selection which could improve the convergence rate of original differential evolution algorithm,accelerated the speed of ELM search parameters,and improved the ELM classification accuracy.Furthermore,it has thesuperior capability of function approximation,good global convergence ability and high optimization precision.The experimental results showed that HKOS-ELM algorithm of added the membership degree preserved the advantages of kernel functions and online learning and improving classification performance of the system.