Welcome to deel-torchlip documentation!

Controlling the Lipschitz constant of a layer or a whole neural network has many applications ranging from adversarial robustness to Wasserstein distance estimation.

This library provides implementation of k-Lispchitz layers for PyTorch.

Content of the library

• k-Lipschitz variant of PyTorch layers such as Linear, Conv2d and AvgPool2d,

• activation functions compatible with pytorch,

• initializers for pytorch,

• loss functions to work with Wasserstein distance estimations.

Installation

You can install deel-torchlip directly from pypi:

pip install deel-torchlip

In order to use deel-torchlip, you also need a valid pytorch installation. deel-torchlip supports torch 1.7.0+.

Cite this work

This library is the PyTorch equivalent of deel-lip, a Tensorflow library built to support the work presented in Achieving robustness in classification using optimal transport with Hinge regularization. This work can be cited as:

@misc{2006.06520,
   Author = {
      Mathieu Serrurier
      and Franck Mamalet
      and Alberto González-Sanz
      and Thibaut Boissin
      and Jean-Michel Loubes
      and Eustasio del Barrio
   },
   Title = {
      Achieving robustness in classification using optimal transport with hinge regularization
   },
   Year = {2020},
   Eprint = {arXiv:2006.06520},
}