trellixvulnteam / project-il-_trkh

Incremental learning

INCREMENTAL LEARNING PROJECT

Marco Ratta and Armando La Rocca

The project is related to the analysis of the incremental learning project. Some different approaches to the problem are studied and some implementations with some possibile improvements are proposed. More details are provided in Report_il.pdf

The implementation of the baselines are provided in ICARL.ipynb and LwF.ipynb and FT.ipynb

The loss with temperature is implemented in icarl_utils.py with some other utilities while cifar100.py and utils.py contains some tools that are useful for the dataset.

Classifier.ipynb is a modified version of Icarl in which are added some methods that allows to use different classifiers.

In ETE.ipynb there is the implentation of the End-to-end approach.

The file NDBC.ipynb is about the new implementation related to the updating of NMC.