Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

A Dynamical System Perspective for Lipschitz Neural Networks

About

The Lipschitz constant of neural networks has been established as a key quantity to enforce the robustness to adversarial examples. In this paper, we tackle the problem of building $1$-Lipschitz Neural Networks. By studying Residual Networks from a continuous time dynamical system perspective, we provide a generic method to build $1$-Lipschitz Neural Networks and show that some previous approaches are special cases of this framework. Then, we extend this reasoning and show that ResNet flows derived from convex potentials define $1$-Lipschitz transformations, that lead us to define the {\em Convex Potential Layer} (CPL). A comprehensive set of experiments on several datasets demonstrates the scalability of our architecture and the benefits as an $\ell_2$-provable defense against adversarial examples.

Laurent Meunier, Blaise Delattre, Alexandre Araujo, Alexandre Allauzen• 2021

Related benchmarks

TaskDatasetResultRank
Certified AccuracyMean Certified Accuracy 255/255
Total Samples (N)121
15
Certified AccuracyMean Certified Accuracy (72/255) (test)
Wins A19
15
Certified RobustnessMean Certified Accuracy 108/255
Wins A15
15
Certified RobustnessMean Certified Accuracy 36/255
Wins A16
15
Mean Accuracy121 benchmarks tasks
Wins A27
15
Image ClassificationCIFAR10 (test)
Accuracy0.785
11
Image ClassificationCIFAR-100 (test)
Clean Accuracy0.478
11
Image ClassificationCIFAR-10 (test)
Clean Accuracy78.5
10
Image ClassificationCIFAR-10 (test)
Clean Accuracy78.5
10
Image ClassificationCIFAR100 (test)
Natural Accuracy0.478
9
Showing 10 of 16 rows

Other info

Follow for update