Metric learning using matrix-based conditional entropy.
-----------------------------------------------------------------------------------
This set of functions reproduces the results of the UCI experiments in the paper:
"Luis G Sanchez Giraldo and Jose Principe, Information Theoretic Learning Using
Infinitely Divisible Kernels. ICLR 2013."
To run the experiments you must:
-- Uncompress "Metric_Learning.tar.gz" to your desired working directory.
-- Download the data.tar.gz from PERSONAL WEBPAGE uncompress in a desired location.
-- Modify the "data_path" variable with the location of the data in your machine.
-- Experiments reported in the above paper can be run by calling:
-- "UCI_comparisons.m"
-- "CEML_UMIS_faces.m"
-- "CEML_alpha_comparison.m"
The function "CondEntropyMetricLearning.m" is a matlab implementation of the metric
learning algorithm described in the paper.
Implementations of NCA, ITML, LMNN, and MCML are part of the Matlab Toolbox for
Dimensionality Reduction by
"Laurens van der Maaten, Delft University of Technology"
This toolbox can be obtained from http://homepage.tudelft.nl/19j49
You are free to use, change, or redistribute this code in any way you want for
non-commercial purposes. However, it is appreciated if you maintain the name of the
original author.
(C) Luis Gonzalo Sanchez Giraldo, 2014