Research On Semi-supervised Incremental Learning Base On Local Sensitive Hash And Support Vector Machine

In recent years,with the development of computer network technology and sensor technology,a large number of high dimensional data have been generated by the application of software and equipment,and the scale is increasing in geometric series.Most of these data are unlabeled samples or samples with a small number of labels and labeling the unlabeled samples is very difficult.At the same time,semi-supervised incremental learning is an important research direction in machine learning,which is of great significance to the development of the information age.In this paper,based on the research of traditional machine learning algorithms,we focus on the semi-supervised incremental learning based on support vector machine.The main research contents are as follows:(1)Access to a large number of domestic and foreign references,discusses the research progress of incremental learning and semi-supervised learning at the present stage and the traditional classifier algorithm cannot adapt to the complex environment,and briefly analyzes the theories of machine learning,semi-supervised learning,support vector machine theory and local sensitive theory.The following research will be expanded on those theories.(2)Because of the shortage of several commonly used incremental learning,we propose a new incremental learning algorithm of SVM with LSH.It uses the LSH algorithm,which can seek similar data fast in a large scale and high dimension data,to filter out the incremental samples which may become SVs with the basis of the SVM algorithm.Then it makes the selected samples and the existing SVs as a basis for the following training.Finally,the experimental results show that this new algorithm can improve the speed of the incremental training learning in large scale data with the effective accuracy.(3)Based on analysis of semi-supervised learning especially for TSVM,we propose a new incremental learning algorithm of TSVM with LSH.It uses the LSH algorithm to filter out and label the first incremental samples which may become similar labeled samples.Then based on TSVM algorithm,it makes the selected samples and the existing SVs which are from last training as a basis for the following semi-supervised incremental training.Finally,experiments show that this new algorithm can improve the speed of the incremental training learning and classification accuracy.Especially in the samples with a relatively small proportion of the label samples,it has the better adaptability.(4)Finally,we combine the proposed learning algorithm with the practical application to show that the algorithm is effective in practice.