Variational Graph Embedding for Globally and Locally Consistent Feature Extraction
published: Oct. 20, 2009, recorded: September 2009, views: 2926
Report a problem or upload filesIf you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.
Existing feature extraction methods explore either global statistical or local geometric information underlying the data. In this paper, we propose a general framework to learn features that account for both types of information based on variational optimization of nonparametric learning criteria. Using mutual information and Bayes error rate as example criteria, we show that high-quality features can be learned from a variational graph embedding procedure, which is solved through an iterative EM-style algorithm where the E-Step learns a variational affinity graph and the M-Step in turn embeds this graph by spectral analysis. The resulting feature learner has several appealing properties such as maximum discrimination, maximum-relevance-minimum-redundancy and locality-preserving. Experiments on benchmark face recognition data sets confirm the effectiveness of our proposed algorithms.
Download slides: ecmlpkdd09_yang_vgeglcfe_01.pdf (689.0 KB)
Download slides: ecmlpkdd09_yang_vgeglcfe_01.ppt (2.3 MB)
Link this pageWould you like to put a link to this lecture on your homepage?
Go ahead! Copy the HTML snippet !