This is the code for the paper:
Ragav Venkatesan, Hemanth Venkateshwara, Sethuraman Panchanathan, Baoxin Li "A strategy for an uncompromising incremental learner"arXiv:1705.00744, 2017.
There are three different codes in this git one for mnist, cifar10 and svhn datasets, each in its own directory. These are the incremental learning experiments. Each directory has a
site_.py
for site Sb and a site_2.py
for Si as mentioned in the paper. site_1.py
will learn both Nb and Gb, each saving its
parameters, confusion matrices and some activities in the records\site_1
directory. These will be loaded when running the site_2.py
which, should be run next.
To run the codes simply do:
.. code-block:: bash
python mnist\site_1.py
python mnist\site_2.py
Run similarly for other datasets also. The directory records
will be created which will hold all results and model parameters, including
layer-wise activities and confusion matrices as described in the paper. All you need will be available in this directory and it is easily
navigable as directories are documented by nomenclature.
The continual learning setups are in mnist-continual and svhn-continual directories and can run the continual learning algorithms. To run these simply run:
.. code-block:: bash
python mnist-continual\continual.py
Results are stored in a similar fashion.
These codes use the yann toolbox internally to run, so that needs to be setup properly.
Thanks for using the code, hope you had fun. Ragav Venkatesan http://www.ragav.net