Structured metric learning for high dimensional problems Export

@inproceedings{citeulike:3662007, abstract = {The success of popular algorithms such as k -means clustering or nearest neighbor searches depend on the assumption that the underlying distance functions reflect domain-specific notions of similarity for the problem at hand. The distance metric learning problem seeks to optimize a distance function subject to constraints that arise from fully-supervised or semisupervised information. Several recent algorithms have been proposed to learn such distance functions in low dimensional settings. One major shortcoming of these methods is their failure to scale to high dimensional problems that are becoming increasingly ubiquitous in modern data mining applications. In this paper, we present metric learning algorithms that scale linearly with dimensionality, permitting efficient optimization, storage, and evaluation of the learned metric. This is achieved through our main technical contribution which provides a framework based on the log-determinant matrix divergence which enables efficient optimization of structured, low-parameter Mahalanobis distances. Experimentally, we evaluate our methods across a variety of high dimensional domains, including text, statistical software analysis, and collaborative filtering, showing that our methods scale to data sets with tens of thousands or more features. We show that our learned metric can achieve excellent quality with respect to various criteria. For example, in the context of metric learning for nearest neighbor classification, we show that our methods achieve 24\% higher accuracy over the baseline distance. Additionally, our methods yield very good precision while providing recall measures up to 20\% higher than other baseline methods such as latent semantic analysis.}, address = {New York, NY, USA}, author = {Davis, Jason V. and Dhillon, Inderjit S.}, booktitle = {KDD '08: Proceeding of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining}, citeulike-article-id = {3662007}, citeulike-linkout-0 = {http://portal.acm.org/citation.cfm?id=1401890.1401918}, citeulike-linkout-1 = {http://dx.doi.org/10.1145/1401890.1401918}, doi = {10.1145/1401890.1401918}, isbn = {978-1-60558-193-4}, keywords = {distance, kdd, low-rank, lsa, lsi, metric-learning}, location = {Las Vegas, Nevada, USA}, pages = {195--203}, posted-at = {2009-03-09 18:08:02}, priority = {0}, publisher = {ACM}, title = {Structured metric learning for high dimensional problems}, url = {http://dx.doi.org/10.1145/1401890.1401918}, year = {2008} }

TY - CONF ID - citeulike:3662007 L3 - citeulike-article-id:3662007 N2 - The success of popular algorithms such as k -means clustering or nearest neighbor searches depend on the assumption that the underlying distance functions reflect domain-specific notions of similarity for the problem at hand. The distance metric learning problem seeks to optimize a distance function subject to constraints that arise from fully-supervised or semisupervised information. Several recent algorithms have been proposed to learn such distance functions in low dimensional settings. One major shortcoming of these methods is their failure to scale to high dimensional problems that are becoming increasingly ubiquitous in modern data mining applications. In this paper, we present metric learning algorithms that scale linearly with dimensionality, permitting efficient optimization, storage, and evaluation of the learned metric. This is achieved through our main technical contribution which provides a framework based on the log-determinant matrix divergence which enables efficient optimization of structured, low-parameter Mahalanobis distances. Experimentally, we evaluate our methods across a variety of high dimensional domains, including text, statistical software analysis, and collaborative filtering, showing that our methods scale to data sets with tens of thousands or more features. We show that our learned metric can achieve excellent quality with respect to various criteria. For example, in the context of metric learning for nearest neighbor classification, we show that our methods achieve 24% higher accuracy over the baseline distance. Additionally, our methods yield very good precision while providing recall measures up to 20% higher than other baseline methods such as latent semantic analysis. CY - Las Vegas, Nevada, USA SN - 978-1-60558-193-4 AD - New York, NY, USA EP - 203 TI - Structured metric learning for high dimensional problems PB - ACM T2 - KDD '08: Proceeding of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining SP - 195 KW - distance KW - kdd KW - low-rank KW - lsa KW - lsi KW - metric-learning AU - Davis, Jason AU - Dhillon, Inderjit PY - 2008/// UR - http://dx.doi.org/10.1145/1401890.1401918 ER -