AI Paper Atlas

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

Synergy-of-Experts: Collaborate to Improve Adversarial Robustness

Sen Cui, Jingfeng Zhang, Jian Liang, Bo Han, Masashi Sugiyama, Changshui Zhang

NeurIPS 2022 · Adversarial robustness and security › Certified and empirical robustness · approach: Robustness-accuracy trade-off analysis · 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
use Synergy-of-Experts collaboration instead of voting ensemble (DVERGE, MoRE, ADP, GAL)SoE with 3 ResNet-20 sub-models · CIFAR10 white-box PGD eps=0.07robust accuracy (%): 12.3 -> 24.0Improved
SoE vs best baseline at small perturbationSoE · CIFAR10 white-box PGD eps=0.01robust accuracy (%): 67.9 -> 72.0Improved
attack SoE with adaptive attacks on dual headsSoE · CIFAR10 white-box PGD eps=0.05robust accuracy (%): 30.4 -> 28.9Hurt
SoE vs baselines under transfer attack with 3 adversarial variantsSoE · CIFAR10 transfer, eps=0.07defense accuracy (%): 69.1 -> 70.2Improved
SoE vs ADP at smallest eps under transfer attackSoE · CIFAR10 transfer, 3 variants, eps=0.01defense accuracy (%): 85.6 -> 85.2No clear effect
SoE vs baselines under transfer attack with 30 adversarial variantsSoE · CIFAR10 transfer, eps=0.07defense accuracy (%): 59.3 -> 61.1Improved
SoE vs ADP at smallest eps with 30 variantsSoE · CIFAR10 transfer, 30 variants, eps=0.01defense accuracy (%): 84.2 -> 83.1Hurt
collaboration training instead of plain ensemble of sub-models (ablation)SoE · CIFAR10 decision boundariesvulnerability areaImproved
increase number of sub-models from 1 to 5 (ablation)SoE · CIFAR10 transfer attacksrobustnessImproved
increase perturbation size epsAll methods · CIFAR10 transfer attacksdefense accuracy (%): 85.2 -> 70.2Hurt

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