AI Paper Atlas

Research map / Adversarial robustness and security / Certified and empirical robustness

Lipschitz-Margin Training: Scalable Certification of Perturbation Invariance for Deep Neural Networks

Yusuke Tsuzuku, Issei Sato, Masashi Sugiyama

NeurIPS 2018 · Adversarial robustness and security › Certified and empirical robustness · approach: Neural network robustness verification · 154 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use Lipschitz-margin training instead of weight decayWide ResNet-16-4 · SVHN test, C&W attack at perturbation sizes 0.2, 0.5, and 1.0accuracy: 72.38 -> 86.90; 20.98 -> 55.11; 2.02 -> 17.69Improved
Use Lipschitz-margin training instead of weight decayWide ResNet-16-4 · SVHN test, cleanaccuracy: 98.31 -> 96.38Hurt
Use Lipschitz-margin training instead of Parseval regularizationWide ResNet-16-4 · SVHN test, C&W attack at perturbation sizes 0.2, 0.5, and 1.0accuracy: 71.35 -> 86.90; 17.92 -> 55.11; 0.94 -> 17.69Improved
Use Lipschitz-margin training instead of Parseval regularizationWide ResNet-16-4 · SVHN test, cleanaccuracy: 98.30 -> 96.38Hurt
Use Lipschitz-margin training instead of spectral norm regularizationWide ResNet-16-4 · SVHN test, C&W attack at perturbation sizes 0.2, 0.5, and 1.0accuracy: 73.66 -> 86.90; 20.35 -> 55.11; 1.39 -> 17.69Improved
Use Lipschitz-margin training instead of spectral norm regularizationWide ResNet-16-4 · SVHN test, cleanaccuracy: 98.27 -> 96.38Hurt
Apply Lipschitz-margin trainingSmall network · Test datamedian certified perturbation size (L2): 0.012 -> 1.02Improved
Replace stride-2, kernel-1 convolution with average-pooling and kernel-1 convolution, and use c=0.1 (ablation)LMT · SVHN testmedian certified perturbation size (L2): larger than 0.08Improved
Replace stride-2, kernel-1 convolution with average-pooling and kernel-1 convolution, and use c=0.1 (ablation)LMT · SVHN testaccuracy: 86%Hurt
Apply the margin addition only when the prediction is correct (ablation)Lipschitz-margin training · Training stabilitytraining stabilityImproved
Increase c linearly to its target value over the first 5 epochs instead of applying strong regularization from the start (ablation)Lipschitz-margin training · Training stabilitytraining stability: 5 epochsImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy