一种基于密度聚类Nystrom抽样算法

A Nystrom Sampling Algorithm Based on Density Clustering

核矩阵在很多机器学习算法中发挥了重要作用,但核矩阵处理的开销非常大。Nystrom方法是流行的抽样方法,抽样使得在处理较大型核矩阵时减少了计算负担。但是,Nystrom方法抽样时采用的是对矩阵进行行、列随机抽样,所以使得准确性受到影响。本文提出了一种基于密度的聚类Nystrom方法,使用密度类算法选出的中心点作为标志点,通过提高聚类的速度和质量来提高Nystrom方法的速度和质量,从而提高了抽样的效率和准确性。

Nuclear matrix has played an important role in many machine learning algorithms, but its calculation is very large. As a popular sampling method, the Nystrom sampleing algorithm reduces the computational burden of dealing with larger nuclear matrix. However, the Nystrom method is based on random sampling from rows or columns of a matrix, affecting the accuracy. The paper presents a Nys- trom method based on density clustering,which employs the algorithm based on density clustering to se- lect a symbol of the center point as landpoints, Therefore, the speed and quality of the Nystrom method can be improved by increasing the speed and quality of clustering, as well the sampling efficiency and ac- curacy will be promoted.