Image Classification on CIFAR10 (train)
100AccuracyAMSGrad
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| AMSGradModel Architecture=CNN (5 conv layers + 1 FC), Batch Size=128, Epochs=1002026.05 | 100 | 0.0001 | — | |
| DiffGradModel Architecture=CNN (5 conv layers + 1 FC), Batch Size=128, Epochs=1002026.05 | 100 | 0 | — | |
| IAdaPID-ADGModel Architecture=CNN (5 conv layers + 1 FC), Batch Size=128, Epochs=1002026.05 | 100 | 0 | — | |
| PGD-ATBackbone=WRN-34-10, Training Time (T/E)=520s2024.03 | 99.9 | — | — | |
| PGD-AT+SLOREBackbone=WRN-34-10, Training Time (T/E)=528s2024.03 | 99.88 | — | — | |
| SL2026.04 | 99.83 | — | — | |
| SL2026.04 | 99.83 | — | — | |
| PGD-AT+LOREBackbone=WRN-34-10, Training Time (T/E)=529s2024.03 | 99.81 | — | — | |
| ACT2026.04 | 99.78 | — | — | |
| PonderNet2026.04 | 99.78 | — | — | |
| ACT2026.04 | 99.78 | — | — | |
| PonderNet2026.04 | 99.78 | — | — | |
| TRADES+LOREBackbone=WRN-34-10, Training Time (T/E)=768s2024.03 | 99.69 | — | — | |
| TRADES+SLOREBackbone=WRN-34-10, Training Time (T/E)=763s2024.03 | 99.65 | — | — | |
| AdaPIDModel Architecture=CNN (5 conv layers + 1 FC), Batch Size=128, Epochs=1002026.05 | 99.51 | 0.006 | — | |
| RIC2026.04 | 99.33 | — | — | |
| RIC2026.04 | 99.33 | — | — | |
| TRADESBackbone=WRN-34-10, Training Time (T/E)=755s2024.03 | 98.98 | — | — | |
| VGG-7-LEBackbone=VGG-7, Augmentation=Local Error (LE), Framework=Equilibrium Propagation (EP)2025.08 | 98.7 | — | — | |
| PGD-ATBackbone=ResNet18, Training Time (T/E)=158s2024.03 | 97.62 | — | — | |
| VGG-7-KDBackbone=VGG-7, Augmentation=Knowledge Distillation (KD), Framework=Equilibrium Propagation (EP)2025.08 | 97.57 | — | — | |
| PGD-AT+LOREBackbone=ResNet18, Training Time (T/E)=161s2024.03 | 97.38 | — | — | |
| PGD-AT+SLOREBackbone=ResNet18, Training Time (T/E)=160s2024.03 | 97.13 | — | — | |
| OursBackbone=ResNet18, Training Data Size=500K2026.07 | 95.71 | — | — | |
| OursBackbone=ResNet18, Training Data Size=400K2026.07 | 95.45 | — | — | |
| TRADES+LOREBackbone=ResNet18, Training Time (T/E)=219s2024.03 | 95.35 | — | — | |
| PGD-AT+LOREBackbone=ResNet50, Training Time (T/E)=185s2024.03 | 95.09 | — | — | |
| RandomBackbone=ResNet18, Training Data Size=500K2026.07 | 95.03 | — | — | |
| RealismBackbone=ResNet18, Training Data Size=500K2026.07 | 94.93 | — | — | |
| OursBackbone=ResNet18, Training Data Size=300K2026.07 | 94.86 | — | — | |
| TRADES+LOREBackbone=ResNet50, Training Time (T/E)=679s2024.03 | 94.83 | — | — | |
| TRADES+SLOREBackbone=ResNet18, Training Time (T/E)=218s2024.03 | 94.77 | — | — | |
| PGD-ATBackbone=ResNet50, Training Time (T/E)=174s2024.03 | 94.77 | — | — | |
| TRADES+SLOREBackbone=ResNet50, Training Time (T/E)=676s2024.03 | 94.74 | — | — | |
| RandomBackbone=ResNet18, Training Data Size=400K2026.07 | 94.68 | — | — | |
| RealismBackbone=ResNet18, Training Data Size=400K2026.07 | 94.64 | — | — | |
| SimilarityBackbone=ResNet18, Training Data Size=500K2026.07 | 94.23 | — | — | |
| RandomBackbone=ResNet18, Training Data Size=300K2026.07 | 93.94 | — | — | |
| RealismBackbone=ResNet18, Training Data Size=300K2026.07 | 93.92 | — | — | |
| TRADESBackbone=ResNet50, Training Time (T/E)=674s2024.03 | 93.89 | — | — | |
| SimilarityBackbone=ResNet18, Training Data Size=400K2026.07 | 93.83 | — | — | |
| PGD-AT+SLOREBackbone=ResNet50, Training Time (T/E)=184s2024.03 | 93.51 | — | — | |
| OursBackbone=ResNet18, Training Data Size=200K2026.07 | 93.43 | — | — | |
| TRADESBackbone=ResNet18, Training Time (T/E)=216s2024.03 | 93.39 | — | — | |
| TRADES(S)+SLOREBackbone=ResNet18, Training Time (T/E)=208s2024.03 | 93.25 | — | — | |
| SimilarityBackbone=ResNet18, Training Data Size=300K2026.07 | 93.21 | — | — | |
| TRADES(S)Backbone=ResNet18, Training Time (T/E)=206s2024.03 | 92.92 | — | — | |
| TRADES(S)+LOREBackbone=ResNet18, Training Time (T/E)=209s2024.03 | 92.79 | — | — | |
| RandomBackbone=ResNet18, Training Data Size=200K2026.07 | 92.78 | — | — | |
| RealismBackbone=ResNet18, Training Data Size=200K2026.07 | 92.6 | — | — | |
| SimilarityBackbone=ResNet18, Training Data Size=200K2026.07 | 91.79 | — | — | |
| MART+SLOREBackbone=WRN-34-10, Training Time (T/E)=374s2024.03 | 91.11 | — | — | |
| MART+LOREBackbone=WRN-34-10, Training Time (T/E)=374s2024.03 | 90.89 | — | — | |
| OursBackbone=ResNet18, Training Data Size=100K2026.07 | 90.47 | — | — | |
| MARTBackbone=WRN-34-10, Training Time (T/E)=369s2024.03 | 90.21 | — | — | |
| RealismBackbone=ResNet18, Training Data Size=100K2026.07 | 89.7 | — | — | |
| OursBackbone=ResNet18, Training Data Size=90K2026.07 | 89.66 | — | — | |
| RandomBackbone=ResNet18, Training Data Size=100K2026.07 | 89.57 | — | — | |
| SimilarityBackbone=ResNet18, Training Data Size=100K2026.07 | 89.41 | — | — | |
| AdamArchitecture=4 x 256 + Skip, Steps (k)=3.2 ± 1.0, Time (s)=30.0 ± 8.9, Time/Step (ms)=9.33 ± 0.202025.06 | 89.2 | — | — | |
| MART+LOREBackbone=ResNet18, Training Time (T/E)=100s2024.03 | 89.04 | — | — | |
| RealismBackbone=ResNet18, Training Data Size=90K2026.07 | 89.03 | — | — | |
| RandomBackbone=ResNet18, Training Data Size=90K2026.07 | 88.73 | — | — | |
| SimilarityBackbone=ResNet18, Training Data Size=90K2026.07 | 88.72 | — | — | |
| ClipBackbone=ResNet18, Training Data Size=500K2026.07 | 88.39 | — | — | |
| MART+LOREBackbone=ResNet50, Training Time (T/E)=214s2024.03 | 88.29 | — | — | |
| MART+SLOREBackbone=ResNet50, Training Time (T/E)=213s2024.03 | 88.25 | — | — | |
| MART+SLOREBackbone=ResNet18, Training Time (T/E)=100s2024.03 | 88.07 | — | — | |
| OursBackbone=ResNet18, Training Data Size=70K2026.07 | 87.9 | — | — | |
| MARTBackbone=ResNet18, Training Time (T/E)=98s2024.03 | 87.85 | — | — | |
| RandomBackbone=ResNet18, Training Data Size=70K2026.07 | 87.74 | — | — | |
| MARTBackbone=ResNet50, Training Time (T/E)=209s2024.03 | 87.66 | — | — | |
| ClipBackbone=ResNet18, Training Data Size=400K2026.07 | 87.56 | — | — | |
| RealismBackbone=ResNet18, Training Data Size=70K2026.07 | 87.45 | — | — | |
| ResNet18#Samples=100002022.10 | 87.31 | — | — | |
| ResNet18#Samples=100002022.10 | 87.31 | — | — | |
| FR-ResNet18#Samples=100002022.10 | 87.2 | — | — | |
| FR-ResNet18#Samples=100002022.10 | 87.2 | — | — | |
| SimilarityBackbone=ResNet18, Training Data Size=70K2026.07 | 87.12 | — | — | |
| ClipBackbone=ResNet18, Training Data Size=300K2026.07 | 86.94 | — | — | |
| ResNet18#Samples=90002022.10 | 86.59 | — | — | |
| ResNet18#Samples=90002022.10 | 86.59 | — | — | |
| FR-ResNet18#Samples=90002022.10 | 86.42 | — | — | |
| FR-ResNet18#Samples=90002022.10 | 86.42 | — | — | |
| FR-ResNet18#Samples=80002022.10 | 85.88 | — | — | |
| FR-ResNet18#Samples=80002022.10 | 85.88 | — | — | |
| RealBackbone=ResNet18, Training Data Size=50K (Original Size)2026.07 | 85.8 | — | — | |
| ClipBackbone=ResNet18, Training Data Size=200K2026.07 | 85.62 | — | — | |
| ResNet18#Samples=80002022.10 | 85.51 | — | — | |
| ResNet18#Samples=80002022.10 | 85.51 | — | — | |
| OursBackbone=ResNet18, Training Data Size=50K (Original Size)2026.07 | 84.96 | — | — | |
| SimilarityBackbone=ResNet18, Training Data Size=50K (Original Size)2026.07 | 84.67 | — | — | |
| RandomBackbone=ResNet18, Training Data Size=50K (Original Size)2026.07 | 84.57 | — | — | |
| FR-ResNet18#Samples=70002022.10 | 84.48 | — | — | |
| FR-ResNet18#Samples=70002022.10 | 84.48 | — | — | |
| ResNet18#Samples=70002022.10 | 84.08 | — | — | |
| ResNet18#Samples=70002022.10 | 84.08 | — | — | |
| RealismBackbone=ResNet18, Training Data Size=50K (Original Size)2026.07 | 83.95 | — | — | |
| FR-ResNet18#Samples=60002022.10 | 83.65 | — | — | |
| FR-ResNet18#Samples=60002022.10 | 83.65 | — | — |