AI Paper Atlas

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

RVCL: Evaluating the Robustness of Contrastive Learning via Verification

Zekai Wang, Weiwei Liu

JMLR 2023 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training with feature consistency

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 contrastive adversarial training strength eps_train from 0 to 8.8/255SimCLR/RoCL ResNet-18 · CIFAR-10 probabilistic verificationACR_CL (x10^-2): 0.832 -> 5.71Improved
increase eps_train from 0 to 8.8/255ResNet-18 · CIFAR-10 PGD robust accuracyrobust accuracy (%): 0.15 -> 22.8Improved
increase eps_train from 0 to 8.8/255ResNet-18 · CIFAR-10 AutoAttack robust accuracyrobust accuracy (%): 0 -> 19.7Improved
increase eps_train from 0 to 8.8/255ResNet-18 · CIFAR-10 clean accuracyclean accuracy (%): 84.8 -> 75.5Hurt
increase eps_train from 0 to 8.8/255ResNet-50 · CIFAR-10 clean accuracyclean accuracy (%): 88.2 -> 79.2Hurt
increase eps_train from 0 to 8.8/255ResNet-18 · CIFAR-100 PGD robust accuracyrobust accuracy (%): 0 -> 12.6Improved
increase eps_train from 0 to 8.8/255ResNet-18 · CIFAR-100 probabilistic verificationACR_CL (x10^-2): 1.60 -> 5.86Improved
use AutoAttack instead of PGD as evaluation attackdeterministic verification models · Mnist/CIFAR-10 robust accuracyrobust accuracy: 5-7% dropHurt
use AutoAttack instead of PGD as evaluation attackprobabilistic verification models · CIFAR-10/100 robust accuracyrobust accuracy: 2-5% dropHurt
adversarially train the downstream layer (eps_trainLE=0.15 vs clean-trained) (ablation)CL encoder with eps_train=0 on Mnist · Mnist robust accuracyrobust accuracy: 0% -> 87.2% (eps_test 0.15 with eps_trainLE 0.05 vs 0.15)Improved
use CBC instead of CROWN verifier for computing ACR_CL (ablation)RVCL · Mnist/CIFAR-10 ACR_CLACR_CLNo clear effect
use CBC instead of CROWN verifier (ablation)RVCL · verification timetime per image: about 300 times slowerHurt

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