jdfr/dynamic_learning_rates_for_continual_unsupervised_learning

Dynamic learning rates for continual unsupervised learning

This is the source code used to get the results in the paper "Dynamic learning rates for continual unsupervised learning":

@article{fernandez2023dynamic,
  title={Dynamic learning rates for continual unsupervised learning},
  author={Fern{\'a}ndez-Rodr{\'\i}guez, Jos{\'e} David and Palomo, Esteban Jos{\'e} and Ortiz-de-Lazcano-Lobato, Juan Miguel and Ramos-Jim{\'e}nez, Gonzalo and L{\'o}pez-Rubio, Ezequiel},
  journal={Integrated Computer-Aided Engineering},
  volume={30},
  number={3},
  pages={257--273},
  year={2023},
  publisher={IOS Press}
}

The code is written by Jose David Fernandez Rodriguez, building upon previous code for competitive clustering techniques (in a classic learning setting, i.e. before devising techniques for continual unsupervised learning) by Esteban Palomo Ferrer and Ezequiel Lopez Rubio. This code is now released under the AGPLv3, as part of a drive to harmonize licensing terms over all my repositories for journal papers.

The code is written as a set of Matlab scripts; it was executed in Matlab 2022a. The main entry points to run experiments are DemoCORA.m and DemoCORACrossValidation.m, while clusteringsCORA.m is used to run classic clustering techniques. The main entry points to crunch data from the experiments are getBestCORAResults.m, showExperimentsBatches.m, showCORAClusters.m and showCORAAndNetwork.m, writeExperimentResults.m, writeExperimentResultsBatches.m, writeExperimentResultsNoBatches.m, and searchGoodEnough.m. A good way to get a sense of the dependencies between the scripts is to use the project analyzer in the Matlab GUI. The file run.py was part of an effort to run minibatch K-means on the dataset used in the paper, as a comparison with our continual learning method.

The dataset used in the paper is available in the cora folder. The images folder contains images used as a dataset to perform an earlier, preliminary set of experiments on continual techniques learning.