Research On Neuron Classification Based On Ensemble Of Extreme Learning Machine

Neuron is the most fundamental structure unit of brain.To understand the structure and function of brain,one has to first investigate the neurons.Since the complex structures and wide varieties of neurons,the classification of neurons is the primary task for further study of neurons.Since the creation of contemporary neuroscience,the classifications of neurons are very controversial.Recent experimental progress accelerates the collection of massive data,which containing the information of morphology,physiology and molecular properties of neuron cells.The massive data encourages researchers to achieve automatic classification of neurons through the powerful machine learning technique.From the morphology aspect,this article studies the classification of neurons to realize the efficient classification of neurons.The detail of works are as follows.(1)Extract the primary morphology characteristics of neurons with Sparse Principal Component Analysis.In the study of neuron classification,when analyzing the morphology characteristics of neuron,we first need to apply the method of dimension reduction because of its complex structures and wide varieties of morphology characteristics.Sparse Principal Component Analysis can efficiently reduce the dimension of high-dimensional data.The primary characteristics after dimension reduction can not only contain most information of the original data,but also help researchers to present analysis of Sparse Principal Component with practical significance since several loadings are zero.(2)Classify neuron cells with Extreme Learning Machine.To classify neuron cells with rapidity and accuracy,we use Extreme Learning Machine to classify the neuron cells,and compared the results with traditional SVM、BP Neural Network algorithms.The experimental results show that Extreme Learning Machine can efficiently classify the neuron cells and exceedSVM and BP Neural Network in both rapidity and accuracy.(3)We propose Regularized Extreme Learning Machine and Ensemble Regularized Extreme Learning Machine based on Adaboost algorithms,and applied in the classification of neurons.To achieve a higher accuracy level of classification,we propose Regularized Online Extreme Learning Machine(ROS-ELM)by introduce the regularization factor into Online Extreme Learning Machine algorithm,and further investigate the influence of regularization factor on the performance of Online Extreme Learning Machine.Based on which,we further proposes a more optimized algorithm,ROS-ELM_Adaboost,by applying Adaoost algorithm to integrate the Regularized Online Extreme Learning Machine.We use ROS-ELM_Adaboost to design the classification model of neuron cells,and verify the advantages of the two algorithms proposed in the paper with comparison test results.