In this project, we propose "Blackout Regularization", a regularization technique for deep learning architectures that supresses connection weights due to penalties assigned by the loss function.
The main purpose of such regularization is to reduce/prevent overfitting.
What things you need to install the software and how to install them
Python 3.6
Tensorflow
For testing, we here use two datasets, namely MNIST and HIGGS dataset. MNIST is downloaded during runtime, HIGGS however, needs to be extracted from the atlas-higgs-challenge-2014-v2.csv.gz to the repository root.
-
Mateusz Garbacz
-
Mike Schoustra
-
Ramy Al Sharif
-
Thomas Puppels
-
Valerie Pourquie
See also the list of contributors who participated in this project.