Online Learning And Its Application For Massive Dataset

With the availability of inexpensive storage and the progress in data capture technology，many organizations have created ultra large databases of business and scientific data， andthis trend is growing continuously. In order to make correct decision， organizations shouldstudy and classify them. If you give up the previous model when new data outcomes， andthen learn all the data again， it will waste a lot of time and space resources. With the size ofdataset increasing， the demand for space and time gradually increased. As a result， learningwill not chase the speed of data updating. For this purpose， a learning system can constantlylearn from new samples from the environment to the new knowledge， and can retain most ofpreviously learned knowledge to become an important means to solve massive data learning，which leads to the algorithm incremental learning to be the urgent needs.In this paper， we combined the existing machine learning algorithms with incrementallearning algorithm to explore new online learning algorithm and have carried on thesimulation experiment. The study mainly includes the following aspects:(1) The paper describes the important research field of machine learningclassification. Track the latest developments of the existing machine learningalgorithms (artificial neural networks，support vector machines， core vectormachines，ball vector machine et)，the existing critical issues and development.After comparing the existing algorithm’s advantages and disadvantages， we getnew challenges facing in this area.(2) An online BVM is presented in this paper: OBVM use the BVM’s advantage thatit can be more efficiently trained with approximate performance. OBVMtransforms a binary classification problem into two single classificationproblems， in which every class is modeled by a hyper-center. Two hyper-centersare incrementally updated by using the same strategy presented in BVM.Moreover， the perpendicular bisector of plane of two hyper-centers is used toclassify data.(3) The online BVM algorithm is applied to the field of gender identification: Firstwe use cascade classifier to locate face， then we use of Active Shape Model offeature points to locate the person’s face. This method not only enhance theaccuracy of the details of the location but also reinforce the realization of robustbackground interference. Then by extracting feature points on the human faceimage to registry image， and then extract the block LBP (local binary pattern)features of the image registration. Finally，we use of the new ball vector machineonline learning algorithm for training and detect gender information of newpictures.