博文
Fast but Not Bad Initial Configuration for Metric MDS
|||
@ARTICLE{XQZZ+12,
AUTHOR = {Xu, Shuo and Qiao, Xiaodong and Zhu, Lijun and Zhang, Yunlinag and Li, Lin},
TITLE = {Fast but Not Bad Initial Configuration for Metric Multidimensional Scaling},
JOURNAL = {Journal of Information & Computational Science},
YEAR = {2012},
volume = {9},
number = {2},
pages = {257--265},
abstract = {The multidimensional scaling (MDS) has become a standard technique
in multivariate data analysis and is widely used in a variety of
disciplines. The objective of MDS is to find a configuration matrix
so that given pairwise dissimilarities can be preserved as
faithfully as possible. But since there exist a lot of non-global
minima for (s)stress, many optimization iterative algorithms are
liable to converge to local minima. Thus, the choice of a good
initial configuration is crucial. Through closer examination on
currently used initial configurations, we find that all these
configurations require extensive preprocessing and usually are
computationally expensive, thus not appropriate for large scale
applications. To overcome this problem, we conjecture that several
approximating scalable solutions for classical MDS can be used to
initialize the configuration for metric MDS at lower complexity, but
should have comparative performance with classical MDS. Finally,
extensive simulation experimental results verify our assumptions.},
keywords = {Multidimensional Scaling (MDS); Classical MDS; (S)stress; FastMap Algorithm; Initial Configuration},
source = {http://www.joics.com/publishedpapers/2012_9_2_257_265.pdf},
}
全文：Xu2012.pdf
https://blog.sciencenet.cn/blog-611051-535707.html
上一篇：四篇应该仔细读的关于文本分析的tutorial类文章
下一篇：利用矩估计Beta分布的参数
