The Research On Incremental Learning Method Based On Density Coverage The Research On Incremental Learning Method Based On Density Coverage

In recent years,with the rapid development of technology,information technology has been continuously exchanged,the data generated has become more and more huge,and the connection between data has become more complicated.People have extracted and extracted useful information from a large amount of data.The more difficult it is.Traditional machine learning can only process these data at one time,and the efficiency of processing constantly updated data is very low,which wastes a lot of time and space.So new method is needed to solve these problems.The covering algorithm is used to build a neural network model from the characteristics of data itself.Essentially,the three-layer neural network model is used to process complex data.The optimization of a neural network problem is transformed into a data coverage problem,which greatly reduces the complexity of the algorithm operation and the space memory occupied by useless data in the system.Therefore,the combination of coverage algorithm and incremental learning method is of great significance to deal with a large number of data which are updated day by day.This paper combines coverage algorithm with incremental learning method,and improves some problems in the combination.The specific work is as follows:1)The coverage algorithm and incremental learning method are studied in detail.An incremental learning model based on density coverage algorithm is proposed by combining the improved density coverage algorithm and incremental learning method of coverage algorithm.Starting with samples with high coverage density,learning paths are planned to obtain the best set of coverage areas from the structure.Then we use density-based coverage algorithm as a new classifier to learn the new data repeatedly,and adjust the classifier through the learning results.In order to improve the basic classifier and optimize the new classifier,we can not only retain the characteristics of previous data,but also learn new data faster and improve the accuracy of learning.Finally,the experimental comparison shows that the efficiency of the improved algorithm is obviously improved compared with the traditional algorithm.2)In incremental learning,due to the dynamic characteristics of data and the updating of a large number of new data,the problem of concept drift arises.This paper presents a mechanism of eliminating forgetfulness to solve this problem.In this paper,the algorithm process is given.Combining with the experimental comparison,the forgetting mechanism of incremental learning of knowledge elimination based on density coverage can solve this problem well.3)Classification of rejected sample points in the process of coverage.This paper analyses the factors affecting the classification of rejection samples,improves the new classification strategy of rejection samples,and makes some experiments to verify the improved strategy.