Research On Improved Algorithms Based On Extreme Learning Machine

In recent years,the intelligent algorithms based on neural network have been widely studied because their application in deep learning,intelligent data processing and large data fields.Among them,extreme learning machine and sparse representation based classification(ELMSRC)has a certain advantage in the data recognition than extreme learning machine;has a certain advantage on model training time than sparse representation based classification;and model training time.However,only use the difference between maximum and second maximum of ELM output vector to select classifier has a certain unreliability in some applications.To solve this problem,we propose the competitive mechanism extreme learning machine and sparse representation classification algorithm(En-SRC),that is,in the stage of classifier selection,the voting based extreme learning machine is used.In addition,in optimization process of weight vector w,self-paced learning with diversity algorithm use voting based extreme learning machine,we propose self-paced extreme learning machine with diversity algorithm;use regularized ELM algorithm,we propose self-paced regularized extreme learning machine with diversity algorithm.The main contributions of this paper are as follows:(1)Proposed the ensemble based extreme learning machine and sparse representation classification algorithm(En-SRC).Extreme learning machine and sparse representation based classification algorithm use extreme learning machine to select classifier.While in the process of dealing with some high noise samples,extreme learning machine is faced with poor results.That is,the reliability of classifier selection is not particularly high.So in the stage of classifier selection,voting based extreme learning machine is adopt rather than extreme learning machine,then it can greatly improve the accuracy of classifier selection.Experimental results show that compared with ELM algorithm,the recognition rate of En-SRC algorithm is increased by 2% to 7%;compared with ELMSRC algorithm,when achieve the same even higher recognition rate,En-SRC algorithm takes less time.(2)Proposed the self-paced extreme learning machine with diversity(SPLD-ELM).Self-paced extreme learning machine with diversity algorithm uses traditional iteration learning algorithm for the weight vector w optimization.While in the process of model training,the iterative methods take a long time.To address this problem,we propose a self-paced extreme learning machine with diversity algorithm,where ELM is adopted to instead of traditional iteration methods for the weight vector w optimization.Experimentalresults show that compared with ELM,the computational complexity of SPLD-ELM algorithm is increased,but the recognition rate is improved effectively.