Regularisation of Neural Networks by Enforcing Lipschitz Continuity

In this video we continue on the topic of Lipschitz continuity by presenting a paper which proposes a projection method to enforce it!

Abstract: We investigate the effect of explicitly enforcing the Lipschitz continuity of neural networks with respect to their inputs. To this end, we provide a simple technique for computing an upper bound to the Lipschitz constant---for multiple p-norms---of a feed forward neural network composed of commonly used layer types. Our technique is then used to formulate training a neural network with a bounded Lipschitz constant as a constrained optimisation problem that can be solved using projected stochastic gradient methods. Our evaluation study shows that the performance of the resulting models exceeds that of models trained with other common regularisers. We also provide evidence that the hyperparameters are intuitive to tune, demonstrate how the choice of norm for computing the Lipschitz constant impacts the resulting model, and show that the performance gains provided by our method are particularly noticeable when only a small amount of training data is available.

[beluga.]

In the paper, they have mentioned that its better to do this for the whole network instead of doing it for each layer in isolation. Any ideas or references on how to do this for the whole network?