List your company

Adversarial Robustness Accuracy on CIFAR-100 (test)

85.21Clean AccuracyAS (adaptive smoothing)
37.21449.674562.13574.5955Apr 1, 2019Mar 5, 2020Feb 8, 2021Jan 14, 2022Dec 20, 2022Nov 25, 2023Oct 30, 2024

Evaluation results

MethodLinks
AS (adaptive smoothing)Robust base model=EDM + TRADES [89]2023.0185.21——————38.72—————————————
RDC2023.0580.47—————————————————53.12——
RDC2024.1080.47—————————————————53.12——
AS (adaptive smoothing)Robust base model=Unlabeled data + TRADES [39]2023.0180.18——————35.15—————————————
DC2023.0579.69—————————————————39.06——
DC2024.1079.69—————————————————39.06——
WRN40-2Backbone=WideResNet-40-22023.0578.13—————————————————0——
WRN40-22024.1078.13—————————————————0——
StandardBackbone=ResNet-182022.0676.721———————22.12252.874———————————
LCNN+GradRegBackbone=ResNet-182022.0676.622———————52.47764.678———————————
Softplus+wt decayBackbone=ResNet-182022.0676.622———————23.90753.766———————————
LCNNBackbone=ResNet-182022.0676.602———————23.70952.179———————————
GradRegBackbone=ResNet-182022.0676.394———————50.29464.678———————————
CUREBackbone=ResNet-182022.0675.928———————50.09663.488———————————
Adv TrainingBackbone=ResNet-182022.0675.521———————55.15566.861———————————
LCNN + GradReg + Adv TrainingBackbone=ResNet-182022.0675.382———————59.22266.861———————————
Wang et al.2023.0575.22—————————————————42.67——
Diffusion (EDM)+TRADES2023.0175.22——————42.67—————————————
AT-EDM2024.1075.22—————————————————42.67——
CURE + GradRegBackbone=ResNet-182022.0674.192———————60.8167.357———————————
Prototype Conformity Loss (PCL)epsilon=4/2552019.0471.949.140.738.6—————————————————
TRADES XCIT-L122023.0170.76——————35.08—————————————
Unlabeled data+TRADES2023.0169.17——————36.98—————————————
3-ensemble of self-ensemblesAdv. train=false, Model=ResNet152, # samples=5122024.0867.71————————————————48.16——37.32
3-ensembleAdv. train=false, Model=ResNet152, # samples=1282024.0866.63————————————————47.66——37.5
Diffusion (DDPM)+AT2023.0165.93——————31.15—————————————
Self-ensembleAdv. train=false, Model=ResNet152, # samples=5122024.0865.71————————————————46.29——30.08
Multi-res backboneAdv. train=false, Model=ResNet152, # samples=1282024.0865.7————————————————25——13.28
CausalDiff2024.1065.62—————————————————56.25——
SCORE Loss AT2023.0165.56——————33.05—————————————
Rebuffi et al.2023.0563.56—————————————————34.64——
Diffusion (DDPM)+TRADES2023.0163.56——————34.64—————————————
AT-DDPM2024.1063.56—————————————————34.64——
RobustResNet-A4#P (M)=147, #F (G)=39.42022.1261.66————34.2530.0429—————————————
Song et al.epsilon=4/2552019.0461.629.327.326.2—————————————————
TRADES2023.0160.86——————30.03—————————————
RobustResNet-A3#P (M)=75.9, #F (G)=19.92022.1260.16————33.5929.5828.48—————————————
AVGTraining Objective=AVG, AdaptiveBudget=None2023.0659.94——————16.1——30.3540.2618.9218.6125.23716.3————
RobustResNet-A2#P (M)=39.0, #F (G)=10.82022.1259.38————3328.7127.68—————————————
RobustResNet-A1#P (M)=19.2, #F (G)=5.112022.1259.34————32.727.7626.75—————————————
WRN-34-R#P (M)=68.1, #F (G)=19.12022.1258.78————31.1727.3326.31—————————————
PDARTSArchitecture=PDARTS2024.0558.4130.35———25.83———————————————
ARNASArchitecture=ARNAS2024.0558.1832.6———29.54———————————————
AdaptiveBudgetTraining Objective=AVG, Budget Norm=l22023.0658.16——————19.5——32.6240.2721.5721.162837.119.7————
AdaptiveBudgetTraining Objective=AVG, Budget Norm=l12023.0657.78——————18.9——33.1641.0321.8121.5528.637.919————
WRN-70-16#P (M)=267, #F (G)=38.82022.1256.93————29.7627.226.12—————————————
WRN-46-14#P (M)=128, #F (G)=18.62022.1256.78————30.0327.2726.28—————————————
AdaptiveBudgetTraining Objective=MAX, Budget Norm=l12023.0656.48——————20.6——29.274025.3424.142335.621.5————
AdvRush (7@96)#P (M)=32.6, #F (G)=4.972022.1256.4————30.426.1625.27—————————————
WRN-28-10#P (M)=36.5, #F (G)=5.202022.1256.3————29.9126.2225.26—————————————
RACL (7@104)#P (M)=32.5, #F (G)=4.932022.1256.09————30.3826.6525.65—————————————
MSDTraining Objective=MSD, AdaptiveBudget=None2023.0656.09——————21——35.540.1424.8325.125.137.621.8————
WRN-34-12#P (M)=66.5, #F (G)=9.602022.1256.08————29.8726.5125.47—————————————
ResNet-18Architecture=ResNet-182024.0555.5726.03———21.44———————————————
AdaptiveBudgetTraining Objective=MAX, Budget Norm=l22023.0655.53——————18.6——29.7839.8525.8724.7620.936.422.3————
AdaptiveBudgetTraining Objective=MSD, Budget Norm=l12023.0655.52——————21.3——30.3140.2826.1925.032435.822.8————
MAXTraining Objective=MAX, AdaptiveBudget=None2023.0655.49——————12.2——25.4539.5525.0321.111336.322————
RobNet-large-v2#P (M)=33.3, #F (G)=5.102022.1255.27————29.2324.6323.69—————————————
AdaptiveBudgetTraining Objective=MSD, Budget Norm=l22023.0654.94——————21.5——28.8739.2825.5924.4324.236.422.7————
DiffPure2023.0539.06—————————————————7.81——
DiffPure2024.1039.06—————————————————7.81——
Adv. w/o DistillAttack=PGD-100, Exit=12023.11—————————————————16.82———
Adv. w/o DistillAttack=PGD-100, Exit=22023.11—————————————————17.24———
Adv. w/o DistillAttack=PGD-100, Exit=32023.11—————————————————19.03———
Adv. w/o DistillAttack=PGD-100, Exit=42023.11—————————————————19.41———
Adv. w/o DistillAttack=PGD-100, Exit=52023.11—————————————————18.97———
Adv. w/o DistillAttack=PGD-100, Exit=62023.11—————————————————17.9———
Adv. w/o DistillAttack=PGD-100, Exit=72023.11—————————————————16.96———
Adv. w/o DistillAttack=PGD-100, Exit=Average2023.11—————————————————18.05———
Adv. w/o DistillAttack=CW, Exit=12023.11—————————————————35.31———
Adv. w/o DistillAttack=CW, Exit=22023.11—————————————————35.72———
Adv. w/o DistillAttack=CW, Exit=32023.11—————————————————36.47———
Adv. w/o DistillAttack=CW, Exit=42023.11—————————————————36.85———
Adv. w/o DistillAttack=CW, Exit=52023.11—————————————————37.2———
Adv. w/o DistillAttack=CW, Exit=62023.11—————————————————37.19———
Adv. w/o DistillAttack=CW, Exit=72023.11—————————————————36.68———
Adv. w/o DistillAttack=CW, Exit=Average2023.11—————————————————36.49———
Adv. w/o DistillAttack=AutoAttack, Exit=12023.11—————————————————31.32———
Adv. w/o DistillAttack=AutoAttack, Exit=22023.11—————————————————34.24———
Adv. w/o DistillAttack=AutoAttack, Exit=32023.11—————————————————37.27———
Adv. w/o DistillAttack=AutoAttack, Exit=42023.11—————————————————39.67———
Adv. w/o DistillAttack=AutoAttack, Exit=52023.11—————————————————41.2———
Adv. w/o DistillAttack=AutoAttack, Exit=62023.11—————————————————41.73———
Adv. w/o DistillAttack=AutoAttack, Exit=72023.11—————————————————40.45———
Adv. w/o DistillAttack=AutoAttack, Exit=Average2023.11—————————————————37.98———
NEO-KDAttack=PGD-100, Exit=12023.11—————————————————20.38———
NEO-KDAttack=PGD-100, Exit=22023.11—————————————————21.29———
NEO-KDAttack=PGD-100, Exit=32023.11—————————————————23.22———
NEO-KDAttack=PGD-100, Exit=42023.11—————————————————24.38———
NEO-KDAttack=PGD-100, Exit=52023.11—————————————————24.44———
NEO-KDAttack=PGD-100, Exit=62023.11—————————————————23.82———
NEO-KDAttack=PGD-100, Exit=72023.11—————————————————22.21———
NEO-KDAttack=PGD-100, Exit=Average2023.11—————————————————22.82———
NEO-KDAttack=CW, Exit=12023.11—————————————————31.56———
NEO-KDAttack=CW, Exit=22023.11—————————————————34.55———
NEO-KDAttack=CW, Exit=32023.11—————————————————38.2———
NEO-KDAttack=CW, Exit=42023.11—————————————————40.6———
NEO-KDAttack=CW, Exit=52023.11—————————————————43.48———
NEO-KDAttack=CW, Exit=62023.11—————————————————44.03———
NEO-KDAttack=CW, Exit=72023.11—————————————————43.01———