基于内核的低秩逼近算法的改进

Improved kernel-based algorithm using low-rank approximation

为进一步提高低秩逼近技术的逼近精度,提出了一种改进的基于内核的低秩逼近算法(IKBLA).算法利用在数值上呈现递减规律的、与矩阵列相关的非均匀概率分布函数对大规模n×n矩阵W进行抽样,接着用抽样得到的小规模c×c矩阵W逼近矩阵W.在UCI数据库中部分数据集上的实验验证了IKBLA的有效性.

In order to further improve the approximation accuracy of low rank approximation technique, an improved kernel-based algorithm using low-rank approximation method （IK- BLA） was proposed in this paper. IKBLA used a matrix columns-dependent non-uniform probability distribution with values subjecting to the law of diminishing to sample the large scale n x n matrix W. Next, approach the matrix W with small scale c x c matrix 17V. Experi- ments in some datasets in UCI database showed the effectiveness of IKBLA.