基于部分失真搜索的自组织映射学习算法

Learning algorithms for self organizing mapping based on partial distortion search

针对传统的自组织映射网络在大数据量或高维情形下训练过程较慢的问题,提出了分别使用部分失真搜索和扩展的部分失真搜索来完成传统算法中最耗时的最近邻搜索过程,减少了完成训练所需乘法次数。实验表明,相对于传统的自组织映射学习算法,所提两种方法分别可以节约近1/3和1/2以上的计算量。

To accelerate the learning process of Self-Organizing Mapping in the situation of large mount of data or high dimension, two learning algorithms were proposed in this paper, by using Partial Distortion Search and Extended Partial Distortion Search respectively to solve the problem of Nearest Neighbor Search during learning process, which could reduce the multiplications greatly. Experiment results indicate that the proposed algorithms can save up to 1/3 and 1/2 multiplications, compared with traditional Self-Organizing Mapping learning algorithm.