The Research Of Extreme Learning Machine Based On Ensemble Learning

Neural networks have been hardly used in real-time computational field for itstime-consuming training and efficiency. In recent years, Extreme LearningMachine(ELM) has been named for a useful tool to solve issues related to training time,which inspires a myriad of researchers to highlight neural networks again. However,when original data blended with noisy data, especially for high dimensional data, therate of classification and regression are largely reduced in ELM, so we will focus on therobustness of ELM later. The robustness of machine learning system can be markedlyimproved by selective ensemble, so we employ the advantage of ensemble learning tomake up the disadvantage of extreme learning machine. This paper is to explore theensemble learning based extreme learning machine and its application inhigh-dimensional blended data.This paper provide three new frameworks including EOP-ELM, AEOP-ELM, andLARSEN-ELM to solve the robustness of ELM, especially for LARSEN-ELM inhigh-dimensional blended data, the algorithm can work well due to several boilerplatereasons: First of all, our algorithm, LARSEN-ELM, employs preprocessing step calledLARS which selects input variables most related to output by criteria, in essence, thisstep can ensure the robustness of our algorithm. Secondly, we consider ELM as thekernel of our algorithm, which speed up our training process in return for time reduction.Finally, the selective ensemble algorithm will definitely select a set of optimal kernelsand then equalize them which ensure the robustness in another perspective. Ouralgorithm, LARSEN-ELM, consists of all steps in above, we explain the improvementof robustness in ELM using LARSEN-ELM theoretically.Finally, this paper, through a large number of MATLAB tests, particularly employsUCI datasets to test the performance of our algorithm with great reliability. From thepoint of view of algorithm, LARSEN-ELM will be a promising direction for futureresearch.