Energy-based Structural Least Squares Twin Parametric-Margin Support Vector Clustering
In few-shot learning,the traditional plane clustering algorithm is sensitive to noise,time-consuming in computation,and the information in the class is not fully utilized,which leads to poor performance of the algorithm.A new energy structured least squares twin parameter-margin support vector clustering for few-shot learning was proposed in this thesis.The algorithm introduced within-class covariance matrix into the objective function to obtain the structure information of data.Furthermore,the energy factor was introduced into the center hyperplane of each cluster parameter-margin center hyperplane,so that the influence of noise and outliers could be reduced in the algorithm,and the optimization problem of the objective function could be solved by using the concave-convex iterative process.Finally,experiments were carried out on multiple synthetic datasets and real datasets,respectively.The significance of the proposed algorithm was verified by statistical tests.The experimental results demonstrated that the proposed algorithm had excellent performance.
万方数据
