ITML: Information Theoretic Metric Learning

About | Downloads | Documentation | Contact

About ITML

ITML is a Matlab implementation of Information Theoretic Metric Learning algorithm. Metric learning involves finding a suitable metric for a given set of data-points with side-information regarding distances between few datapoints. ITML characterizes the metric using a Mahalanobis distance function and learns the associated parameters using Bregman's cyclic projection algorithm.

This software is currently being actively maintained; please check back often for updates. When using this code, please cite ITML and the relevant paper.

@inproceedings{davis-et-al-icml-2007,
  author    = {Jason V. Davis and
               Brian Kulis and
               Prateek Jain and
               Suvrit Sra and
               Inderjit S. Dhillon},
  title     = {Information-theoretic metric learning},
  booktitle = {ICML},
  year      = {2007},
  pages     = {209-216},
  address   = {Corvalis, Oregon, USA}
  month     = {June}
}

@manual{itml,
  title        = {Information Theoretic Metric Learning},
  author       = {Jason V. Davis and Brian Kulis and Prateek Jain and Suvrit Sra and Inderjit S. Dhillon},
  organization = {UT, Austin},
  address      = {http://www.cs.utexas.edu/users/pjain/itml/},
}

Downloads

Source code

Download the latest version (1.2, released 2008-4-28): itml-1.2.tar.gz

Download version 1.1, released 2008-2-11): itml-1.1.tar.gz

Example code and data

• Iris Data

Documentation