Skip to content
This repo contains active learning query strategies as introduced in our GCPR 2013 paper.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
queryFunctions
roc-analysis
LICENSE
README.md
active_learning_gpml.m
auroc.m
covmin.m
demo.m
initWorkspace.m
min_kernel.m

README.md

COPYRIGHT

This package contains Matlab source code for active learning experiments with Gaussian process regression models as described in:

Alexander Freytag and Erik Rodner and Paul Bodesheim and Joachim Denzler: "Labeling examples that matter: Relevance-Based Active Learning with Gaussian Processes". Proceedings of the German Conference on Pattern Recognition (GCPR), 2013.

Please cite that paper if you are using this code!

(LGPL) copyright by Alexander Freytag and Erik Rodner and Paul Bodesheim and Joachim Denzler

CONTENT

initWorkspace.m demo.m active_learning_gpml.m queryFunctions/select_sample_random.m queryFunctions/select_sample_gpMean.m queryFunctions/select_sample_gpVariance.m queryFunctions/select_sample_gpUncertainty.m queryFunctions/select_sample_gpWeight.m queryFunctions/select_sample_gpImpact.m roc-analysis/roc.m roc-analysis/auroc.m README.txt License.txt

USAGE

First download the GPML (see below) and change the gpml path in initWorkspace.m

  • run initWorkspace

  • getting to know:
    (1) run demo.m, which performs an AL experiment on synthetically created 5D-histograms. Change the settings in the demo file to figure out how everything works. Don't forget to set up the gpml directory accordingly!

  • run an AL experiment: (1) Use the function "active_learning_gpml" to run a comparison of different AL-strategies for a given scenario (2) Please refer to the documentation in active_learning_gpml.m for explanations of input and output variables.

NOTE

To keep things simple, GP models are computed using the GPML toolbox:

http://mloss.org/revision/download/1206/

@MISC{Rasmussen10:GPML, author = {C. E. {Rasmussen} and H. {Nickisch}}, title = {GPML Gaussian Processes for Machine Learning Toolbox}, year = {2010}, note = {\url{http://mloss.org/software/view/263/}}, }

For computing kernel values needed within the experiments, you can rely on the GPML toolbox as well.

You can’t perform that action at this time.