/lipschitz-neural-networks

Code for running experiments described in "Regularisation of Neural Networks by Enforcing Lipschitz Continuity"

Primary LanguageD

Lipschitz Continuous Neural Networks

This repo contains the code used to run the experiments for an extended version of Gouk et al. (2018). The code uses dopt, a deep learning framework written in D.

Example

The following command will train a wide residual network on the CIFAR-10 dataset:

./cifar10.d --datapath ~/Datasets/cifar10/ --norm=inf --lambda=3 --arch=wrn

References

Gouk, H., Frank, E., Pfahringer, B., & Cree, M. (2018). Regularisation of Neural Networks by Enforcing Lipschitz Continuity. arXiv preprint arXiv:1804.04368.