Image Classification on CIFAR-100 (Adversarial Robustness PGD/AA)
80.73Clean AccuracyStandard Training
Evaluation Results
| Method | Links | ||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Standard TrainingType=Standard2025.05 | 80.73 | 0 | 0 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FractalDBBackbone=WideResNet-40-4, Data Augmentation Pipeline=PixMix2026.03 | 80 | 1.5 | — | — | 65 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Fractal artsBackbone=WideResNet-40-4, Data Augmentation Pipeline=PixMix2026.03 | 79.7 | 6.8 | — | — | 66.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FVisBackbone=WideResNet-40-4, Data Augmentation Pipeline=PixMix2026.03 | 79 | 7.7 | — | — | 69.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MoireDB (Ours, Static)Backbone=WideResNet-40-4, Data Augmentation Pipeline=PixMix2026.03 | 78.9 | 6.5 | — | — | 69.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BaselineBackbone=WideResNet-40-42026.03 | 78.7 | 3.2 | — | — | 50 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| VisualAtomBackbone=WideResNet-40-4, Data Augmentation Pipeline=PixMix2026.03 | 78.6 | 1.5 | — | — | 66.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TRADESModel=ResNet-342026.06 | 77.2 | — | — | — | — | — | — | — | — | — | — | 25.6 | — | 6.6 | 18 | 29.3 | 47.5 | 25.9 | 21.5 | 30.7 | — | — | |
| TaFDModel=ResNet-342026.06 | 76.9 | — | — | — | — | — | — | — | — | — | — | 31.9 | — | 13.6 | — | 37.4 | 51.2 | 33.4 | 19.8 | 34.2 | — | — | |
| BenignArchitecture=WideResNet-282026.05 | 76.78 | — | 0 | — | — | — | — | — | — | — | 0.01 | — | 7.24 | — | — | — | — | — | — | — | — | — | |
| FACEModel=ResNet-342026.06 | 74.5 | — | — | — | — | — | — | — | — | — | — | 1.6 | — | 0 | 0 | 0.3 | 9.5 | 0.8 | 0 | 0.4 | — | — | |
| BenignArchitecture=ResNet-182026.05 | 74.08 | — | 0 | — | — | — | — | — | — | — | 0.01 | — | 5.72 | — | — | — | — | — | — | — | — | — | |
| BenignArchitecture=PreActResNet-182026.05 | 73.89 | — | 0 | — | — | — | — | — | — | — | 0.01 | — | 4.81 | — | — | — | — | — | — | — | — | — | |
| Mean-ATModel=ResNet-342026.06 | 73.5 | — | — | — | — | — | — | — | — | — | — | 25.6 | — | 7.3 | 24.3 | 30.3 | 44.9 | 26.2 | 16.7 | 29.8 | — | — | |
| Consistency-AT+DUCATepsilon=8/2552026.03 | 72.29 | 30.73 | 25.66 | — | — | 49.95 | 30.3 | 19.16 | 14.21 | 11.45 | — | — | — | — | — | — | — | — | — | — | — | — | |
| MNGModel=ResNet-342026.06 | 71.7 | — | — | — | — | — | — | — | — | — | — | 22.6 | — | 4.7 | 16.4 | 30.3 | 43.6 | 25.3 | 8.5 | 29.4 | — | — | |
| TaFDModel=MobileViT-XS2026.06 | 71.4 | — | — | — | — | — | — | — | — | — | — | 26.6 | — | 22.4 | 41.1 | 21.6 | 39.2 | 19.3 | 21.8 | 21 | — | — | |
| PGD-AT+DUCATepsilon=8/2552026.03 | 70.71 | 29.56 | 25.2 | — | — | 48.6 | 28.54 | 16.56 | 10.55 | 14.65 | — | — | — | — | — | — | — | — | — | — | — | — | |
| EB-JDAT-SADAJEMbase_model=SADAJEM2025.05 | 68.32 | 38.42 | 35.57 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ATASArchitecture=WideResNet-282026.05 | 68.18 | — | 7.21 | — | — | — | — | — | — | — | 16.07 | — | 40.85 | — | — | — | — | — | — | — | — | — | |
| EB-JDAT-JEM++base_model=JEM++2025.05 | 68.05 | 37.21 | 35.19 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FACEModel=MobileViT-XS2026.06 | 67.9 | — | — | — | — | — | — | — | — | — | — | 1.5 | — | 0 | 0.1 | 0.2 | 8.9 | 0.8 | 0 | 0.2 | — | — | |
| TRADESModel=MobileViT-XS2026.06 | 66.8 | — | — | — | — | — | — | — | — | — | — | 15.5 | — | 11.2 | 37.3 | 12.5 | 23.2 | 9.7 | 4.7 | 9.9 | — | — | |
| GBNModel=ResNet-342026.06 | 65 | — | — | — | — | — | — | — | — | — | — | 9.7 | — | 0 | 0.2 | 15.7 | 21.2 | 11.2 | 2.7 | 16.8 | — | — | |
| LAS-AWP2025.05 | 64.89 | 36.36 | 30.77 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| 6-layer-CNN-SLBackbone=6-layer CNN, Training=Supervised Learning2026.06 | 64.75 | — | — | 2.53 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CAAT-LoRABackbone=ViT-B, Tuned params (M)=1.09, Tuned / Total (%)=1.282026.04 | 64.01 | — | 24.52 | — | — | 25.14 | — | — | — | — | 28.05 | 25.9 | — | — | — | — | — | — | — | — | — | — | |
| HYPERATBackbone=ViT-B, Tuned params (M)=5.47, Tuned / Total (%)=6.422026.04 | 63.93 | — | 23.64 | — | — | 24.63 | — | — | — | — | 25.9 | 24.72 | — | — | — | — | — | — | — | — | — | — | |
| LoRA-16Backbone=ViT-B, Tuned params (M)=1.24, Tuned / Total (%)=1.462026.04 | 63.85 | — | 22.13 | — | — | 24.56 | — | — | — | — | 25.79 | 24.16 | — | — | — | — | — | — | — | — | — | — | |
| 6-layer-CNN-SL-advBackbone=6-layer CNN, Training=Adversarial Supervised Learning2026.06 | 63.61 | — | — | 2.83 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ATASArchitecture=PreActResNet-182026.05 | 63.57 | — | 17.17 | — | — | — | — | — | — | — | 21.1 | — | 27.34 | — | — | — | — | — | — | — | — | — | |
| ATASArchitecture=ResNet-182026.05 | 63.39 | — | 17.5 | — | — | — | — | — | — | — | 21.68 | — | 27.71 | — | — | — | — | — | — | — | — | — | |
| CAAT-ADAPTERBackbone=ViT-B, Tuned params (M)=1.07, Tuned / Total (%)=1.262026.04 | 63.15 | — | 23.87 | — | — | 24.86 | — | — | — | — | 26.47 | 25.07 | — | — | — | — | — | — | — | — | — | — | |
| PGD-2Architecture=WideResNet-282026.05 | 62.96 | — | 22.29 | — | — | — | — | — | — | — | 26.23 | — | 30.62 | — | — | — | — | — | — | — | — | — | |
| FULLBackbone=ViT-B, Tuned params (M)=85.22, Tuned / Total (%)=1002026.04 | 62.01 | — | 27.5 | — | — | 29.15 | — | — | — | — | 30.56 | 29.07 | — | — | — | — | — | — | — | — | — | — | |
| DHAT-CFA2025.05 | 61.54 | 37.15 | 30.93 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ADAPTER-32Backbone=ViT-B, Tuned params (M)=1.22, Tuned / Total (%)=1.432026.04 | 61.17 | — | 20.36 | — | — | 20.88 | — | — | — | — | 22.79 | 21.34 | — | — | — | — | — | — | — | — | — | — | |
| FULLLORA-ATBackbone=ViT-B, Tuned params (M)=2.46, Tuned / Total (%)=2.892026.04 | 61.12 | — | 23.75 | — | — | 24.77 | — | — | — | — | 26.01 | 24.84 | — | — | — | — | — | — | — | — | — | — | |
| SGLR2025.05 | 61.02 | 32.98 | 28.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RiFTBackbone=ResNet-18, Training Strategy=TRADES2026.06 | 60.73 | 28.51 | 23.91 | — | — | 25.42 | — | — | — | — | — | — | 31.74 | — | — | — | — | — | — | — | 27.92 | 24.95 | |
| Mean-ATModel=MobileViT-XS2026.06 | 60.7 | — | — | — | — | — | — | — | — | — | — | 17.3 | — | 17.8 | 40.1 | 13 | 21.8 | 9.8 | 8.8 | 10 | — | — | |
| GradAlignArchitecture=WideResNet-282026.05 | 60.43 | — | 7.74 | — | — | — | — | — | — | — | 9.96 | — | 41.21 | — | — | — | — | — | — | — | — | — | |
| RiFTBackbone=ResNet-18, Training Strategy=AT2026.06 | 60.17 | 26.08 | 22.75 | — | — | 25.26 | — | — | — | — | — | — | 30.43 | — | — | — | — | — | — | — | 25.34 | 23.89 | |
| MultiGradArchitecture=WideResNet-282026.05 | 60.07 | — | 23.08 | — | — | — | — | — | — | — | 27.37 | — | 31.02 | — | — | — | — | — | — | — | — | — | |
| FPCMBackbone=ResNet-18, Training Strategy=AT2026.06 | 60.05 | 28.49 | 24.25 | — | — | 27.05 | — | — | — | — | — | — | 32.24 | — | — | — | — | — | — | — | 27.45 | 25.38 | |
| ADAPTER-8Backbone=ViT-B, Tuned params (M)=1.07, Tuned / Total (%)=1.262026.04 | 60.01 | — | 18.63 | — | — | 19.57 | — | — | — | — | 21.35 | 19.85 | — | — | — | — | — | — | — | — | — | — | |
| 6-layer-CNN-RLBackbone=6-layer CNN, Training=Reinforcement Learning2026.06 | 59.8 | — | — | 13.06 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EMFFBackbone=ResNet-18, Training Strategy=TRADES2026.06 | 59.35 | 29.26 | 24.49 | — | — | 25.68 | — | — | — | — | — | — | 32.43 | — | — | — | — | — | — | — | 28.79 | 25.28 | |
| ATBackbone=ResNet-18, Training Strategy=AT2026.06 | 59.26 | 25.17 | 21.26 | — | — | 24.28 | — | — | — | — | — | — | 29.83 | — | — | — | — | — | — | — | 24.16 | 22.85 | |
| TRADESBackbone=ResNet-18, Training Strategy=TRADES2026.06 | 59.02 | 27.44 | 23.05 | — | — | 24.42 | — | — | — | — | — | — | 31.29 | — | — | — | — | — | — | — | 26.98 | 24.07 | |
| FPCMBackbone=ResNet-18, Training Strategy=TRADES2026.06 | 58.98 | 29.4 | 24.5 | — | — | 26.03 | — | — | — | — | — | — | 31.95 | — | — | — | — | — | — | — | 29.05 | 25.45 | |
| A3Backbone=ResNet-18, Training Strategy=MART2026.06 | 58.86 | 29.79 | 24.24 | — | — | 27.28 | — | — | — | — | — | — | 33.01 | — | — | — | — | — | — | — | 28.7 | 26.01 | |
| FSRBackbone=ResNet-18, Training Strategy=AT2026.06 | 58.84 | 25.24 | 21.94 | — | — | 24.51 | — | — | — | — | — | — | 29.1 | — | — | — | — | — | — | — | 24.36 | 23.05 | |
| AURORABackbone=ViT-B, Tuned params (M)=1.15, Tuned / Total (%)=1.352026.04 | 58.67 | — | 21.55 | — | — | 23.44 | — | — | — | — | 24.03 | 23.01 | — | — | — | — | — | — | — | — | — | — | |
| PGD-2Architecture=PreActResNet-182026.05 | 58.65 | — | 20.72 | — | — | — | — | — | — | — | 24.96 | — | 28.6 | — | — | — | — | — | — | — | — | — | |
| LBGAT2025.05 | 58.64 | 32.75 | 28.33 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SORAArchitecture=WideResNet-282026.05 | 58.58 | — | 23.77 | — | — | — | — | — | — | — | 27.93 | — | 31.23 | — | — | — | — | — | — | — | — | — | |
| CASBackbone=ResNet-18, Training Strategy=AT2026.06 | 58.48 | 24.83 | 20.12 | — | — | 24.82 | — | — | — | — | — | — | 28.92 | — | — | — | — | — | — | — | 23.84 | 22.55 | |
| PGD-2Architecture=ResNet-182026.05 | 58.41 | — | 20.65 | — | — | — | — | — | — | — | 25.01 | — | 28.39 | — | — | — | — | — | — | — | — | — | |
| AGR-TRADES2025.05 | 58.17 | 28.34 | 25.84 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PGD-10Architecture=WideResNet-282026.05 | 57.97 | — | 25.2 | — | — | — | — | — | — | — | 29.57 | — | 32.26 | — | — | — | — | — | — | — | — | — | |
| AWP2025.05 | 57.94 | 33.75 | 28.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ELLEArchitecture=PreActResNet-182026.05 | 57.94 | — | 19.7 | — | — | — | — | — | — | — | 23.79 | — | 27.74 | — | — | — | — | — | — | — | — | — | |
| MARTBackbone=ResNet-18, Training Strategy=MART2026.06 | 57.92 | 27.41 | 22.19 | — | — | 24.96 | — | — | — | — | — | — | 31.36 | — | — | — | — | — | — | — | 26.67 | 24.38 | |
| MNGModel=MobileViT-XS2026.06 | 57.9 | — | — | — | — | — | — | — | — | — | — | 17.9 | — | 19.1 | 39.8 | 13.9 | 21.5 | 10.2 | 10 | 10.6 | — | — | |
| FTA2CBackbone=ResNet-18, Training Strategy=TRADES2026.06 | 57.88 | 28.31 | 24 | — | — | 25.66 | — | — | — | — | — | — | 31.47 | — | — | — | — | — | — | — | 27.91 | 24.91 | |
| FGSMArchitecture=WideResNet-282026.05 | 57.83 | — | 0 | — | — | — | — | — | — | — | 0.02 | — | 49.82 | — | — | — | — | — | — | — | — | — | |
| TRADESArchitecture=WideResNet-282026.05 | 57.76 | — | 24.85 | — | — | — | — | — | — | — | 30.15 | — | 32.17 | — | — | — | — | — | — | — | — | — | |
| CIFSBackbone=ResNet-18, Training Strategy=TRADES2026.06 | 57.68 | 27.52 | 22.19 | — | — | 24.48 | — | — | — | — | — | — | 30.3 | — | — | — | — | — | — | — | 26.94 | 23.97 | |
| UIAT2025.05 | 57.65 | 33.91 | 29.03 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FSRBackbone=ResNet-18, Training Strategy=TRADES2026.06 | 57.63 | 27.57 | 23.35 | — | — | 24.53 | — | — | — | — | — | — | 30.39 | — | — | — | — | — | — | — | 27.01 | 24.13 | |
| ELLEArchitecture=ResNet-182026.05 | 57.53 | — | 19.69 | — | — | — | — | — | — | — | 23.94 | — | 27.68 | — | — | — | — | — | — | — | — | — | |
| GradAlignArchitecture=PreActResNet-182026.05 | 57.5 | — | 20.67 | — | — | — | — | — | — | — | 25.2 | — | 28.62 | — | — | — | — | — | — | — | — | — | |
| A3Backbone=ResNet-18, Training Strategy=AT2026.06 | 57.5 | 32.34 | 24.36 | — | — | 27.44 | — | — | — | — | — | — | 33.89 | — | — | — | — | — | — | — | 31.78 | 26.52 | |
| CASBackbone=ResNet-18, Training Strategy=TRADES2026.06 | 57.41 | 27.41 | 22.08 | — | — | 24.9 | — | — | — | — | — | — | 30.22 | — | — | — | — | — | — | — | 26.59 | 24.26 | |
| A3Backbone=ResNet-18, Training Strategy=TRADES2026.06 | 57.36 | 29.61 | 24.81 | — | — | 26.35 | — | — | — | — | — | — | 31.71 | — | — | — | — | — | — | — | 28.81 | 25.8 | |
| TDAT2025.05 | 57.32 | 33.17 | 26.61 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| N-FGSMArchitecture=WideResNet-282026.05 | 57.28 | — | 23.74 | — | — | — | — | — | — | — | 27.82 | — | 30.5 | — | — | — | — | — | — | — | — | — | |
| GradAlignArchitecture=ResNet-182026.05 | 57.24 | — | 20.97 | — | — | — | — | — | — | — | 25.37 | — | 28.85 | — | — | — | — | — | — | — | — | — | |
| EMFFBackbone=ResNet-18, Training Strategy=AT2026.06 | 57.17 | 27.79 | 24.02 | — | — | 26.4 | — | — | — | — | — | — | 31.49 | — | — | — | — | — | — | — | 27.02 | 25.22 | |
| AAERArchitecture=WideResNet-282026.05 | 57.1 | — | 23.75 | — | — | — | — | — | — | — | 27.92 | — | 30.54 | — | — | — | — | — | — | — | — | — | |
| MART+DUCATepsilon=8/2552026.03 | 56.73 | 34.32 | 27.44 | — | — | 29.66 | 29.44 | 20.92 | 17.33 | 10.11 | — | — | — | — | — | — | — | — | — | — | — | — | |
| CIFSBackbone=ResNet-18, Training Strategy=AT2026.06 | 56.64 | 25.14 | 20.41 | — | — | 24.35 | — | — | — | — | — | — | 29.11 | — | — | — | — | — | — | — | 24.21 | 22.98 | |
| 6-layer-CNN-RL-advBackbone=6-layer CNN, Training=Adversarial Reinforcement Learning2026.06 | 56.54 | — | — | 25.51 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| FSRBackbone=ResNet-18, Training Strategy=MART2026.06 | 56.45 | 27.01 | 23.02 | — | — | 25.27 | — | — | — | — | — | — | 30.55 | — | — | — | — | — | — | — | 26.28 | 24.15 | |
| ZeroGradArchitecture=PreActResNet-182026.05 | 56.28 | — | 20.81 | — | — | — | — | — | — | — | 25.32 | — | 28.37 | — | — | — | — | — | — | — | — | — | |
| ZeroGradArchitecture=ResNet-182026.05 | 56.22 | — | 20.96 | — | — | — | — | — | — | — | 25.49 | — | 28.6 | — | — | — | — | — | — | — | — | — | |
| RiFTBackbone=ResNet-18, Training Strategy=MART2026.06 | 56.15 | 29.25 | 23.92 | — | — | 25.96 | — | — | — | — | — | — | 32.32 | — | — | — | — | — | — | — | 28.78 | 25.26 | |
| FTA2CBackbone=ResNet-18, Training Strategy=AT2026.06 | 56.12 | 26.69 | 21.82 | — | — | 24.85 | — | — | — | — | — | — | 30.75 | — | — | — | — | — | — | — | 26.02 | 23.32 | |
| FGSMArchitecture=PreActResNet-182026.05 | 55.97 | — | 0 | — | — | — | — | — | — | — | 0.16 | — | 63.93 | — | — | — | — | — | — | — | — | — | |
| EMFFBackbone=ResNet-18, Training Strategy=MART2026.06 | 55.87 | 28.45 | 24.14 | — | — | 25.81 | — | — | — | — | — | — | 31.47 | — | — | — | — | — | — | — | 27.88 | 25.21 | |
| MultiGradArchitecture=PreActResNet-182026.05 | 55.53 | — | 21.45 | — | — | — | — | — | — | — | 26.09 | — | 28.98 | — | — | — | — | — | — | — | — | — | |
| MultiGradArchitecture=ResNet-182026.05 | 55.48 | — | 21.61 | — | — | — | — | — | — | — | 26.17 | — | 29.17 | — | — | — | — | — | — | — | — | — | |
| TRADESArchitecture=PreActResNet-182026.05 | 55.42 | — | 22.77 | — | — | — | — | — | — | — | 28.05 | — | 29.86 | — | — | — | — | — | — | — | — | — | |
| FTA2CBackbone=ResNet-18, Training Strategy=MART2026.06 | 55.38 | 27.45 | 21.8 | — | — | 24.21 | — | — | — | — | — | — | 30.85 | — | — | — | — | — | — | — | 26.65 | 23.27 | |
| CASBackbone=ResNet-18, Training Strategy=MART2026.06 | 55.24 | 26.68 | 21.12 | — | — | 25.11 | — | — | — | — | — | — | 30.27 | — | — | — | — | — | — | — | 25.89 | 23.31 | |
| FPCMBackbone=ResNet-18, Training Strategy=MART2026.06 | 55.08 | 25.5 | 21.24 | — | — | 22.54 | — | — | — | — | — | — | 28.14 | — | — | — | — | — | — | — | 24.94 | 22.19 | |
| TRADESArchitecture=ResNet-182026.05 | 55 | — | 22.79 | — | — | — | — | — | — | — | 28.16 | — | 30.13 | — | — | — | — | — | — | — | — | — | |
| SORAArchitecture=PreActResNet-182026.05 | 54.74 | — | 21.56 | — | — | — | — | — | — | — | 26.12 | — | 28.89 | — | — | — | — | — | — | — | — | — | |
| MART2025.05 | 54.69 | 31.9 | 27.25 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — |