Nonparametric Divergence Estimation

Liang Xiong, Dougal Sutherland, Barnabás Póczos, Jeff Schneider

Point of contact: Liang Xiong and Dougal J. Sutherland

Software Information


These two packages implement the non-parametric divergence estimators described in:
Barnabas Poczos, Liang Xiong, Jeff Schneider,
Nonparametric divergence estimation with applications to machine learning on distributions
Uncertainty in Artificial Intelligence, 2011

The C++ implementation is faster and can exploit many-core hardware more easily, but takes more work to install. You can download version 0.1 here or check out the most recent source on github. It includes a file-based Matlab wrapper with a similar interface to that of the pure Matlab package, as well as a standalone command-line interface and a shared library to call its functions directly.

The pure-Matlab version is slower (by a factor of about 10-15) and is limited to computing with eight cores with the Parallel Computation Toolbox, but is easier to set up.

NEW The Python version is now also available (divergence computation only), featuring the speed close to C and the ease of use close to Matlab.

To see how these estimators work in the real world image classification tasks, check out our paper and full experiment code here.


divergence estimation, group anomaly, non-parametric


Download (text/plain, 388 bytes)

Copyright 2010, Carnegie Mellon University, Auton Lab. All Rights Reserved.