Regularisation of Neural Networks by Enforcing Lipschitz Continuity

TL;DR

This paper introduces a simple method to enforce Lipschitz continuity in neural networks, improving model performance especially with limited data by formulating it as a constrained optimization problem.

Contribution

It provides a straightforward technique to compute Lipschitz bounds for neural networks and integrates this into training as a constrained optimization, outperforming common regularizers.

Findings

Models with enforced Lipschitz continuity outperform those with standard regularizers.

The method is effective with small training datasets.

Hyperparameters are intuitive to tune for the proposed method.

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.