Research On Data Classification Methods Based On Non-local Operators On The Graph

With the development of information technology,the amount of data has exploded.How to classify these data in order to achieve efficient use,which is one of the important content of the current data mining.Among them,variational methods based on high-dimensional data classification of sparse graphs are becoming important research directions in the field of machine learning.This method is mainly based on the idea of graph cut,by constructing a non-directional weighted graph to achieve the classification of data.In the framework of discrete non-local total variances,this dissertation integrates equality constraints(two kinds),one-way inequality constraints,two-way inequality constraints,Ratio Cut,Normalized Cut and Cheeger Cut to systematically analyze the performance of different balance constraints.They are seven different methods of constraint.In the process of studying different constraints,Normalized Cut was found that the constraint ability of balance is not enough,especially when the data set is not balanced,the constraint ability will be further reduced.In order to solve this problem,Normalized Cut is improved to introduce the degree equilibrium constraint into the energy functional and improve the constraint ability of Normalized Cut.In the classification of many kinds of high-dimensional data,computational efficiency is low because the traditional Potts model classifies n-class data by n label functions.In order to improve the computational efficiency,this paper proposes a vectorization acceleration method based on Potts model,and simplifies the algorithm by the concept of parallel computation.In order to simplify the calculation of the above model,Augmented Lagrangian Method is used in this paper,and the projection method of constraint processing is used in each step to reduce the number of Lagrange multipliers and the number of penalty parameters.In order to validate the validity of the proposed algorithm in a more scientific and effective way,this paper uses multiple experiments of standard data sets and artificial data sets in the field of international data classification respectively.And taking the average value of experimental results.The accuracy and efficiency of the proposed model and the fast algorithm are compared,and the validity of the proposed model and algorithm is verified.