摘要
流形学习有助于发现数据的内在分布和几何结构.目前已有的流形学习算法对噪音和算法参数都比较敏感,噪音使得输入参数更加难以选择,参数较小的变化会导致差异显著的学习结果.针对Isomap这一流形学习算法,提出了一种新方法,通过引入集成学习技术,扩大了可以产生有效可视化结果的输入参数范围,并且降低了对噪音的敏感性.
Manifold learning is helpful to the discovery of the intrinsic distribution and geometry structure of data. Current manifold learning algorithms are usually sensitive to noise and input parameters. The appearance of noise and the change of input parameters usually produce significantly different learning results. In this paper, a new method is proposed based on the manifold learning algorithm Isomap through introducing ensemble learning technique, which enlarges the value range that the input parameters can take to generate good visualization effect and reduces the sensitivity to noise.
出处
《计算机研究与发展》
EI
CSCD
北大核心
2005年第9期1533-1537,共5页
Journal of Computer Research and Development
基金
国家杰出青年科学基金项目(60325207)
教育部优秀青年教师基金项目
霍英东基金项目(91067)~~
关键词
机器学习
流形学习
集成学习
可视化
machine learning
manifold learning
ensemble learning
visualization
