AI Paper Atlas

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

Phase Transition from Clean Training to Adversarial Training

Yue Xing, Qifan Song, Guang Cheng

NeurIPS 2022 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training recipes · 0 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 attack strength eps from 0 to 0.5/255Adversarial training PreActResNet18 · CIFAR-10adversarial testing loss: 0.1837 -> 0.2937Hurt
Increase attack strength eps from 0.5/255 to 2/255Adversarial training · CIFAR-10adversarial testing accuracy: 0.8977 -> 0.8101Hurt
Increase eps beyond eps* (0.5, then 2, 8)Adversarial training · CIFAR-10 connectivitylow-loss region overlap with clean modelHurt
Use FGM instead of PGD at eps below eps*FGSM adversarial training · CIFAR-10catastrophic overfitting: 0.5/255 vs 8/255No clear effect
Reduce training size from 50K to 10KAdversarial training PreActResNet18 · CIFAR-10eps*: 0.5/255 -> 3.3/255Improved
Reduce training size from 50K to 25KAdversarial training PreActResNet18 · CIFAR-100eps*: 1.0/255 -> 4.0/255Improved
Widen MNIST hidden layer from 16 to 128 nodesMNIST CNN adversarial training · MNISTeps*: 26.1/255 -> 48.0/255Improved
Widen WideResNet34 from width 1 to 10Adversarial training · CIFAR-10eps*: 1.1/255 -> 1.1/255No clear effect
Replace vanilla adversarial loss with TRADESPreActResNet18 · CIFAR-10eps*: 0.5/255 -> 0.7/255Improved
Replace vanilla adversarial loss with MARTPreActResNet18 · CIFAR-10eps*: 0.5/255 -> 0.8/255Improved
Estimate eps* by adversarial fine-tuning with PGD, 1 to 3 runsExtrapolation algorithm · CIFAR-10 n=10000 Linfeps* estimate (true 3.3/255): 4.7/255 -> 2.5/255Mixed
Estimate eps* with FGM 3 runs instead of PGD 3 runsExtrapolation algorithm · CIFAR-10 n=50000 Linfeps* estimate (true 0.5/255): 0.52/255 -> 0.59/255Hurt

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