List your company
ResearchFormerly WizWand

Image Classification on CIFAR-10 standard (test) (Standard Accuracy)

98.09AccuracySAT
4.354828.689953.02577.3601Nov 7, 2020Sep 22, 2021Aug 8, 2022Jun 24, 2023May 8, 2024Mar 24, 2025Feb 7, 2026
Updated 8mo ago

Evaluation results

MethodLinks
SATBackbone=ResNet-50, Pre-training Dataset=ImageNet-1K, Pre-training Data=entire set, Running time (hours)=286.1, Finetuning Mode=SFF (standard full finetuning)2023.0298.09
SAT with RCSBackbone=ResNet-50, Subset fraction k=0.2, Running time (hours)=79.8, Finetuning Mode=SFF (standard full finetuning)2023.0298.06
SAT with RCSBackbone=ResNet-50, Subset fraction k=0.1, Running time (hours)=55.4, Finetuning Mode=SFF (standard full finetuning)2023.0297.82
SAT with RCSBackbone=ResNet-50, Subset fraction k=0.05, Running time (hours)=48.2, Finetuning Mode=SFF (standard full finetuning)2023.0297.65
Fast-ATBackbone=ResNet-50, Pre-training Dataset=ImageNet-1K, Pre-training Data=entire set, Running time (hours)=10.4, Finetuning Mode=SFF (standard full finetuning)2023.0297.54
Standard trainingBackbone=ResNet-50, Pre-training Dataset=ImageNet-1K, Pre-training Data=entire set, Finetuning Mode=SFF (standard full finetuning)2023.0297.41
STAA-SNN (Ours)Architecture=ResNet-19, Timestep=42025.0397.14
STAA-SNN (Ours)Architecture=ResNet-19, Timestep=22025.0396.85
STAA-SNN (Ours)Architecture=ResNet-19, Timestep=12025.0396.75
MPBNArchitecture=ResNet-19, Timestep=42025.0396.52
MPBNArchitecture=ResNet-19, Timestep=22025.0396.47
SAT with RandomBackbone=ResNet-50, Subset fraction k=0.2, Running time (hours)=70.3, Finetuning Mode=SFF (standard full finetuning)2023.0296.1
MPBNArchitecture=ResNet-19, Timestep=12025.0396.06
PFAArchitecture=ResNet-19, Timestep=42025.0395.71
SAT with RandomBackbone=ResNet-50, Subset fraction k=0.1, Running time (hours)=45.8, Finetuning Mode=SFF (standard full finetuning)2023.0295.6
PFAArchitecture=ResNet-19, Timestep=22025.0395.6
Spike-driven TransformerArchitecture=Transformer-2-512, Timestep=42025.0395.6
IM-LIFArchitecture=ResNet-19, Timestep=32025.0395.29
SAT with RandomBackbone=ResNet-50, Subset fraction k=0.05, Running time (hours)=38.7, Finetuning Mode=SFF (standard full finetuning)2023.0295.27
STAA-SNN (Ours)Architecture=VGG-13, Timestep=42025.0395.26
LSGArchitecture=ResNet-19, Timestep=42025.0395.17
STAA-SNN (Ours)Architecture=ResNet-20, Timestep=42025.0395.03
GLIFArchitecture=ResNet-19, Timestep=42025.0394.85
SpikingformerArchitecture=Spikingformer-4-25, Timestep=42025.0394.77
STAA-SNN (Ours)Architecture=VGG-13, Timestep=22025.0394.7
GLIFArchitecture=ResNet-19, Timestep=22025.0394.44
TETArchitecture=ResNet-19, Timestep=42025.0394.44
LSGArchitecture=ResNet-19, Timestep=22025.0394.41
Standard TrainingBackbone=RN-18, Training Data=Clean, Mitigation Protocol=None2023.0594.37
STAA-SNN (Ours)Architecture=ResNet-20, Timestep=22025.0394.35
MPBNArchitecture=ResNet-20, Timestep=42025.0394.28
TETArchitecture=ResNet-19, Timestep=22025.0394.16
SpikformerArchitecture=Spikformer-4-25, Timestep=42025.0393.94
Diet-SNNArchitecture=VGG-16, Timestep=52025.0393.85
IM-lossArchitecture=VGG-16, Timestep=52025.0393.85
MPBNArchitecture=ResNet-20, Timestep=22025.0393.54
SATBackbone=ResNet-50, Pre-training Dataset=ImageNet-1K, Pre-training Data=entire set, Running time (hours)=286.1, Finetuning Mode=SLF (standard linear finetuning)2023.0293.53
SAT with RCSBackbone=ResNet-50, Subset fraction k=0.2, Running time (hours)=79.8, Finetuning Mode=SLF (standard linear finetuning)2023.0293.48
STAA-SNN (Ours)Architecture=ResNet-20, Timestep=12025.0393.08
SAT with RCSBackbone=ResNet-50, Subset fraction k=0.1, Running time (hours)=55.4, Finetuning Mode=SLF (standard linear finetuning)2023.0292.92
SAT with RCSBackbone=ResNet-50, Subset fraction k=0.05, Running time (hours)=48.2, Finetuning Mode=SLF (standard linear finetuning)2023.0292.68
PearsonNoise Type=Uniform, Noise Level=Low, Bias Correction=false2020.1192.37
MPBNArchitecture=ResNet-20, Timestep=12025.0392.22
Diet-SNNArchitecture=ResNet-20, Timestep=52025.0391.78
Fast-ATBackbone=ResNet-50, Pre-training Dataset=ImageNet-1K, Pre-training Data=entire set, Running time (hours)=10.4, Finetuning Mode=SLF (standard linear finetuning)2023.0290.91
Orthogonal ProjectionBackbone=RN-18, Training Data=Random Noise (class-wise linearly separable), Mitigation Protocol=Ortho Proj2023.0590.37
Orthogonal ProjectionBackbone=RN-18, Training Data=Clean, Mitigation Protocol=Ortho Proj2023.0590.16
Orthogonal ProjectionBackbone=RN-18, Training Data=Unlearnable Examples [10] (class-wise linearly separable), Mitigation Protocol=Ortho Proj2023.0589.98
Orthogonal ProjectionBackbone=RN-18, Training Data=LSP [36], Mitigation Protocol=Ortho Proj2023.0587.99
Orthogonal ProjectionBackbone=RN-18, Training Data=OPS [35] (class-wise linearly separable), Mitigation Protocol=Ortho Proj2023.0587.94
SAT with RandomBackbone=ResNet-50, Subset fraction k=0.2, Running time (hours)=70.3, Finetuning Mode=SLF (standard linear finetuning)2023.0287.69
Adversarial TrainingBackbone=RN-18, Training Data=Clean, Mitigation Protocol=Adv Training2023.0587.16
SAT with RandomBackbone=ResNet-50, Subset fraction k=0.1, Running time (hours)=45.8, Finetuning Mode=SLF (standard linear finetuning)2023.0287.14
Adversarial TrainingBackbone=RN-18, Training Data=Robust Unlearnable [28], Mitigation Protocol=Adv Training2023.0587.07
Orthogonal ProjectionBackbone=RN-18, Training Data=Regions-4 [26] (class-wise linearly separable), Mitigation Protocol=Ortho Proj2023.0586.87
Adversarial TrainingBackbone=RN-18, Training Data=Random Noise (class-wise linearly separable), Mitigation Protocol=Adv Training2023.0586.36
LogicTreeNet-GModel=CNN, Gate count=61.0 M2026.0286.29
Adversarial TrainingBackbone=RN-18, Training Data=LSP [36], Mitigation Protocol=Adv Training2023.0586.18
Adversarial TrainingBackbone=RN-18, Training Data=Regions-4 [26] (class-wise linearly separable), Mitigation Protocol=Adv Training2023.0585.74
SAT with RandomBackbone=ResNet-50, Subset fraction k=0.05, Running time (hours)=38.7, Finetuning Mode=SLF (standard linear finetuning)2023.0285.72
Adversarial TrainingBackbone=RN-18, Training Data=Unlearnable Examples [10], Mitigation Protocol=Adv Training2023.0585.64
Adversarial TrainingBackbone=RN-18, Training Data=Unlearnable Examples [10] (class-wise linearly separable), Mitigation Protocol=Adv Training2023.0585.56
Adversarial TrainingBackbone=RN-18, Training Data=Adversarial Poisoning [7], Mitigation Protocol=Adv Training2023.0585.32
Adversarial TrainingBackbone=RN-18, Training Data=NTGA [37], Mitigation Protocol=Adv Training2023.0584.85
Adversarial TrainingBackbone=RN-18, Training Data=AR (l2) [27], Mitigation Protocol=Adv Training2023.0584.09
JeffreyNoise Type=Uniform, Noise Level=High, Bias Correction=false2020.1183.8
Orthogonal ProjectionBackbone=RN-18, Training Data=NTGA [37], Mitigation Protocol=Ortho Proj2023.0582.21
Standard trainingBackbone=ResNet-50, Pre-training Dataset=ImageNet-1K, Pre-training Data=entire set, Finetuning Mode=SLF (standard linear finetuning)2023.0278.84
BitLogicModel=FFN, Gate count=384 K2026.0272.36
LogicTreeNet-MModel=CNN, Gate count=3.08 M2026.0271.01
Orthogonal ProjectionBackbone=RN-18, Training Data=Unlearnable Examples [10], Mitigation Protocol=Ortho Proj2023.0565.17
LILogicNet-LModel=FFN, Gate count=256 K2026.0260.98
DiffLogic Net-LModel=FFN, Gate count=1.28 M2026.0260.78
LogicTreeNet-SModel=CNN, Gate count=400 K2026.0260.38
DMINoise Type=Random, Noise Level=0.7, Bias Correction=false2020.1157.81
LILogicNet-MModel=FFN, Gate count=64 K2026.0257.66
DiffLogic Net-MModel=FFN, Gate count=512 K2026.0257.39
LILogicNet-SModel=FFN, Gate count=8 K2026.0255.11
DiffLogic Net-SModel=FFN, Gate count=48 K2026.0251.27
BitLogicModel=CNN, Gate count=253.4 K2026.0250.53
Standard TrainingBackbone=RN-18, Training Data=NTGA [37], Mitigation Protocol=None2023.0540.78
Orthogonal ProjectionBackbone=RN-18, Training Data=OPS+EM [35], Mitigation Protocol=Ortho Proj2023.0539.43
Standard TrainingBackbone=RN-18, Training Data=Robust Unlearnable [28], Mitigation Protocol=None2023.0526.86
Orthogonal ProjectionBackbone=RN-18, Training Data=Robust Unlearnable [28], Mitigation Protocol=Ortho Proj2023.0525.83
Standard TrainingBackbone=RN-18, Training Data=LSP [36], Mitigation Protocol=None2023.0525.75
Standard TrainingBackbone=RN-18, Training Data=Unlearnable Examples [10], Mitigation Protocol=None2023.0524.56
Standard TrainingBackbone=RN-18, Training Data=OPS+EM [35], Mitigation Protocol=None2023.0520.11
Standard TrainingBackbone=RN-18, Training Data=OPS [35] (class-wise linearly separable), Mitigation Protocol=None2023.0515.35
Orthogonal ProjectionBackbone=RN-18, Training Data=Adversarial Poisoning [7], Mitigation Protocol=Ortho Proj2023.0514.74
Standard TrainingBackbone=RN-18, Training Data=Unlearnable Examples [10] (class-wise linearly separable), Mitigation Protocol=None2023.0514.49
Standard TrainingBackbone=RN-18, Training Data=AR (l2) [27], Mitigation Protocol=None2023.0513.77
Orthogonal ProjectionBackbone=RN-18, Training Data=AR (l2) [27], Mitigation Protocol=Ortho Proj2023.0513.03
Adversarial TrainingBackbone=RN-18, Training Data=OPS+EM [35], Mitigation Protocol=Adv Training2023.0512.22
Adversarial TrainingBackbone=RN-18, Training Data=OPS [35] (class-wise linearly separable), Mitigation Protocol=Adv Training2023.0511.77
Standard TrainingBackbone=RN-18, Training Data=Regions-4 [26] (class-wise linearly separable), Mitigation Protocol=None2023.0510.32
Standard TrainingBackbone=RN-18, Training Data=Random Noise (class-wise linearly separable), Mitigation Protocol=None2023.0510.03
Standard TrainingBackbone=RN-18, Training Data=Adversarial Poisoning [7], Mitigation Protocol=None2023.057.96