AI Paper Atlas

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

Adversarial Contrastive Learning via Asymmetric InfoNCE

Qiying Yu, Jieming Lou, Xianyuan Zhan, Qizhang Li, Wangmeng Zuo, Yang Liu, Jingjing Liu

ECCV 2022 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training with feature consistency · 25 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 RoCL loss with inferior-positives loss L_IPRoCL · CIFAR-10 Linear Probingstandard accuracy: 83.84 -> 87.63Improved
replace RoCL loss with L_IP+HNRoCL · CIFAR-10 Linear Probingrobust accuracy: 38.98 -> 42.96Improved
replace RoCL loss with L_IPRoCL · CIFAR-10 Adversarial Full Finetuningrobust accuracy: 50.54 -> 50.29No clear effect
replace RoCL loss with L_IP+HNRoCL · CIFAR-100 Linear Probingrobust accuracy: 18.49 -> 22.54Improved
replace AdvCL loss with L_IP+HNAdvCL · CIFAR-10 Linear Probingrobust accuracy: 51.00 -> 52.65Improved
replace AdvCL loss with L_IP+HNAdvCL · CIFAR-10 Adversarial Linear Finetuningrobust accuracy: 52.38 -> 53.19No clear effect
replace AdvCL loss with L_HNAdvCL · CIFAR-100 Adversarial Linear Finetuningstandard accuracy: 47.45 -> 47.32Hurt
add hard negatives L_HN to AdvCL, transfer CIFAR-100 to STL-10AdvCL · CIFAR-100 to STL-10 Adversarial Linear Finetuningrobust accuracy: 32.14 -> 33.62Improved
use L_IP+HN for AdvCL under AutoAttackAdvCL · CIFAR-100 Linear Probing AutoAttackrobust accuracy: 19.57 -> 21.43Improved
use L_IP+HN instead of AdvCL under CIFAR-10 AutoAttack linear finetuningAdvCL · CIFAR-10 Adversarial Linear Finetuning AutoAttackrobust accuracy: 43.93 -> 43.01Hurt
use fixed alpha instead of symmetric (alpha=0.5) (ablation)CoreACL · CIFAR-10standard accuracy: 78.90 -> 81.29Improved
add adaptive alpha annealing (ablation)CoreACL · CIFAR-10robust accuracy: 50.27 -> 51.37Improved

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