AI Paper Atlas

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

Robust Mixture-of-Expert Training for Convolutional Neural Networks

Yihua Zhang, Ruisi Cai, Tianlong Chen, Guanhua Zhang, Huan Zhang, Pin-Yu Chen, Shiyu Chang, Zhangyang Wang, Sijia Liu

ICCV 2023 · Adversarial robustness and security › Certified and empirical robustness · approach: Robust overfitting and regularization · 40 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
train MoE-CNN with router-expert alternating bi-level AdvMoE instead of standard ATMoE-CNN ResNet-18 · CIFAR-10 (PGD robust accuracy)RA-PGD (%): 45.57 -> 51.83Improved
AdvMoE (r=0.5 pathway) versus dense model with twice the widthWRN-28-10 · CIFAR-10 (PGD robust accuracy)RA-PGD (%): 51.75 -> 55.73Improved
AdvMoE versus dense baseline on clean accuracy with ResNet-18MoE-CNN ResNet-18 · CIFAR-10 (clean accuracy)SA (%): 82.99 -> 80.15Hurt
use MoE-CNN pathway instead of dense modelAdvMoE · CIFAR-10 (inference cost)GFLOPS: 0.54 -> 0.15Improved
evaluate with AutoAttack instead of PGDAdvMoE · CIFAR-10 (AutoAttack)RA-AA (%): 40.42 -> 45.13Improved
increase number of experts N to 10 (with r = 0.8) (ablation)AdvMoE · CIFAR-10 ResNet-18 expert count sweepRAHurt
use small per-expert scale r<0.5 with N>=8 (ablation)AdvMoE · CIFAR-10 ResNet-18 expert count sweepRAHurt
increase model scale r from 0.2 to 0.8 (ablation)AdvMoE · CIFAR-10RAImproved
raise training PGD steps from 2 to 10 (ablation)AdvMoE · CIFAR-10 ResNet-18RA (%): 51.83 -> 52.22No clear effect
apply AdvMoE to MoE-ViTDeiT-Tiny MoE · ImageNet MoE-ViTRA (%): 44.63 -> 45.93Improved

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