AI Paper Atlas

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

Improved Adversarial Training via Learned Optimizer

Yuanhao Xiong, Cho-Jui Hsieh

ECCV 2020 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training refinements · 23 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
replace constant-step PGD inner maximizer with backtracking line search (ablation)AdvTrain · MNISTrobust accuracy (%): 87.33 -> 88.71Improved
attack with PGD+BLS instead of PGD (ablation)AdvTrain model · MNISTrobust accuracy (%): 87.33 -> 86.09Improved
replace PGD inner maximizer with learned RNN optimizerAdvTrain · MNISTmin robust accuracy (%): 94.28 -> 95.51Improved
replace PGD inner maximizer with learned RNN optimizerAdvTrain · CIFAR-10 (VGG-16)min robust accuracy (%): 40.75 -> 42.89Improved
replace PGD inner maximizer with learned RNN optimizerAdvTrain · CIFAR-10 (WideResNet)min robust accuracy (%): 45.13 -> 45.98Improved
replace PGD inner maximizer with learned RNN optimizerTRADES · CIFAR-10 (VGG-16)min robust accuracy (%): 45.95 -> 47.23Improved
replace PGD inner maximizer with learned RNN optimizerTRADES · CIFAR-10 (WideResNet)min robust accuracy (%): 52.68 -> 54.11Improved
replace PGD inner maximizer with learned RNN optimizerTRADES · MNISTmin robust accuracy (%): 95.50 -> 95.80Improved
replace PGD inner maximizer with learned RNN optimizerTRADES · Restricted ImageNet (ResNet-18)min robust accuracy (%): 4.80 -> 5.79Improved
use RNN optimizer rather than CNN-based L2LDA attackerL2LDA · CIFAR-10 (VGG-16)min robust accuracy (%): 34.27 -> 47.23Improved
extend learned RNN attacker from 10 to 40 stepsRNN-TRADES · CIFAR-10 (VGG-16)robust accuracy (%): 49.49 -> 49.28Improved
replace PGD training with learned RNN optimizer on natural accuracyTRADES · CIFAR-10 (WideResNet)natural accuracy (%): 85.89 -> 84.21Hurt

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