Nonlinear metric learning with kernel density estimation

NCU Institutional Repository > 總教學中心委員會 > 語言中心 > 期刊論文 > Item 987654321/107868

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: https://ir.lib.ncu.edu.tw/handle/987654321/107868

题名:	Nonlinear metric learning with kernel density estimation
作者:	陳雯玲;He, Yujie;Mao, Yi;Chen, Wenlin;Chen, Yixin
贡献者:	總教學中心語言中心
关键词:	Algorithm design and analysis;Algorithms;Classification;Density;Density measurement;Euclidean distance;Kernel;kernel density estimation;Kernels;large margin nearest neighbors;Learning;Learning systems;Mapping;Mathematical analysis;metric learning;neighborhood components analysis;Nonlinearity;Optimization algorithms;Vectors
日期:	2015-06-01
上传时间:	2026-04-23 14:27:38 (UTC+8)
出版者:	IEEE Computer Society;New York: IEEE
摘要:	摘要： Metric learning, the task of learning a good distance metric, is a key problem in machine learning with ample applications. This paper introduces a novel framework for nonlinear metric learning, called kernel density metric learning (KDML), which is easy to use and provides nonlinear, probability-based distance measures. KDML constructs a direct nonlinear mapping from the original input space into a feature space based on kernel density estimation. The nonlinear mapping in KDML embodies established distance measures between probability density functions, and leads to accurate classification on datasets for which existing linear metric learning methods would fail. It addresses the severe challenge to distance-based classifiers when features are from heterogeneous domains and, as a result, the Euclidean or Mahalanobis distance between original feature vectors is not meaningful. We also propose two ways to determine the kernel bandwidths, including an adaptive local scaling approach and an integrated optimization algorithm that learns the Mahalanobis matrix and kernel bandwidths together. KDML is a general framework that can be combined with any existing metric learning algorithm. As concrete examples, we combine KDML with two leading metric learning algorithms, large margin nearest neighbors (LMNN) and neighborhood component analysis (NCA). KDML can naturally handle not only numerical features, but also categorical ones, which is rarely found in previous metric learning algorithms. Extensive experimental results on various datasets show that KDML significantly improves existing metric learning algorithms in terms of classification accuracy. 其他題名： TKDE 出版者： New York: IEEE 出版日期： 2015-06-01 出處： IEEE transactions on knowledge and data engineering, 2015-06, Vol.27 (6), p.1602-1614 資源來源： IEEE Xplore Digital Library (LUT) 版權： Copyright The Institute of Electrical and Electronics Engineers, Inc. (IEEE) Jun 2015 識別號： ISSN: 1041-4347 識別號： EISSN: 1558-2191 識別號： DOI: 10.1109/TKDE.2014.2384522 識別號： CODEN: ITKEEH
显示于类别:	[語言中心] 期刊論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	37	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....