Research On Machine Learning Algorithms With ELM And NMF

Extreme learning machine (ELM) is a competitive machine learning technique,which is simple in theory and fast in implementation; it can identify faults quickly andprecisely as compared with traditional identification techniques such as support vectormachines (SVM). As verified by the simulation results, ELM tends to have betterscalability and can achieve much better generalization performance. Therefore, it iswidely used in various fields.AdaBoost is one of the most popular algorithms of classifier ensemble toimprove the generalization performance.The ordinary algorithm of classifier aim tosovle the problem of two classes, such as SVM. Although there exists somealgorithms which aim to distinguish more than two classes, they perform not wellbecause of two much calculation or power ability of recognition.However, until now, not so much works have been done to apply AdaBoost toELM for multiclass classification problem directly. In view of this, this paperproposes a structure to apply ELM and MAELM to Local Binary Patterns (LBP)based face recognition problem. Experiments in LBP based face recognition willshow that the proposed algorithm outperforms the original ELM.On the other side, with the rapid development of information technology, a largenumber of high-dimensional data appear constantly.The increasement ofhigh-dimensional data has the challenge for machine learning. The nonnegativeMatrix decomposition (NMF), as a powerful data dimension reduction method inmachine learning, is widely used in many aspects.After learning various effective algorithm of NMF,this paper propose a new wayin which we add two constraints(orthogonal not identical,ONI and Graph RegularizedNonnegative Matrix Factorization，GNMF).And we prove it in theory and realityNMF combined with Extreme Learning Machine (ELM) feature mappingmethod (EFM NMF),proposed by Qing He,can effectively reduce the computationalof the NMF.In view of this theory, we also combine the proposed algorithm withELM and make some tests in practice. The results show that this method greatly improves the performance of the algorithm in different hidden nodes and has a certainreference value.