AI Paper Atlas

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

Which Models have Perceptually-Aligned Gradients? An Explanation via Off-Manifold Robustness

Suraj Srinivas, Sebastian Bordt, Himabindu Lakkaraju

NeurIPS 2023 · Adversarial robustness and security › Certified and empirical robustness · approach: Decision-boundary geometry of robustness · 1 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
increase robustness regularization (gradient norm, randomized smoothing, smoothness penalty) from none to moderateResnet18 · CIFAR-10off-manifold sensitivityImproved
increase robustness regularization from none to moderateResnet18 · CIFAR-10on-manifold sensitivityNo clear effect
increase robustness regularization beyond the intermediate regime (excessive robustness)Resnet18 · CIFAR-10accuracyHurt
increase robustness regularization from weak to intermediateResnet18 input gradients · CIFAR-10LPIPS perceptual similarity to scoreImproved
increase robustness regularization beyond the peakResnet18 input gradients · CIFAR-10LPIPS perceptual similarity to scoreHurt
train excessively robust PGD modelsResnet18 · ImageNet-64x64accuracy and perceptual alignmentHurt
use l2-adversarially robust training instead of standard trainingResnet18 · MNIST with distractorrelative noise robustness (distractor vs signal)Improved
use adversarially robust pretrained modelsResnet18 · ImageNetLPIPS perceptual similarity to scoreImproved

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