List your company
ResearchFormerly WizWand

Image Classification on CIFAR-10 4,000 labels (test)

2.7Test Error RateUDA (RandAugment)
0.96412.68224.436.118Jul 9, 2015May 1, 2017Feb 23, 2019Dec 17, 2020Oct 11, 2022Aug 4, 2024May 29, 2026
Updated 5mo ago

Evaluation results

MethodLinks
UDA (RandAugment)Model=PyramidNet, # Param=26M2019.042.7—
100% SupervisedBackbone=Shake-Shake2019.082.86—
UDA (RandAugment)Model=Shake-Shake, # Param=26M2019.043.7—
FixMatch + IAM-Dlabeled samples=40002026.053.88—
FixMatchlabeled samples=40002026.054.1—
FixMatch + CMW-Net2022.024.25—
FixMatch2022.024.26—
UDA (RandAugment)Model=WRN-28-2, # Param=1.5M2019.044.32—
UDA2022.024.88—
MixMatchNumber of labeled samples=4000, Model parameters=26M2019.054.95—
MixMatchModel=WRN, # Param=26M2019.044.95—
SWANumber of labeled samples=4000, Model parameters=26M2019.055—
Fast-SWAModel=Shake-Shake, # Param=26M2019.045—
HybridNetBackbone=Shake-Shake2019.086.09—
MixMatchModel=WRN-28-2, # Param=1.5M2019.046.24—
Mean TeacherNumber of labeled samples=4000, Model parameters=26M2019.056.28—
Mean TeacherModel=Shake-Shake, # Param=26M2019.046.28—
MixMatch2022.026.42—
ICTModel=Conv-Large, # Param=3.1M2019.047.29—
ICTModel=WRN-28-2, # Param=1.5M2019.047.66—
Mean Teacher2022.029.19—
Fast-SWA-480Classifier=13-layer CNN (similar), Data Augmentation=Standard2019.129.86—
Shake-Shake (26 2x96d) + AutoAugmentBackbone=Shake-Shake (26 2x96d), Augmentation=AutoAugment2018.0510—
Triple-GAN-V2Classifier=13-layer CNN [32], Data Augmentation=Standard2019.1210.01—
SWA-480Classifier=13-layer CNN (similar), Data Augmentation=Standard2019.1210.3—
VAT + EntMinModel=Conv-Large, # Param=3.1M2019.0410.55—
Label PropagationClassifier=13-layer CNN (similar), Data Augmentation=Standard2019.1210.61—
SNTGClassifier=13-layer CNN (similar), Data Augmentation=Standard2019.1210.93—
SNTGModel=Conv-Large, # Param=3.1M2019.0410.93—
VATClassifier=13-layer CNN (similar), Data Augmentation=Standard2019.1211.36—
LGA + VATModel=WRN-28-2, # Param=1.5M2019.0412.06—
MTClassifier=13-layer CNN (similar), Data Augmentation=Standard2019.1212.31—
Mean TeacherModel=Conv-Large, # Param=3.1M2019.0412.31—
Π-ModelModel=Conv-Large, # Param=3.1M2019.0412.36—
VATlabels_count=4000, data_augmentation=None2018.0913.15—
MT (our code base)Classifier=13-layer CNN [32], Data Augmentation=Standard2019.1213.18—
II modelClassifier=13-layer CNN (similar), Data Augmentation=Standard2019.1213.2—
Shake-Shake (26 2x96d)Backbone=Shake-Shake (26 2x96d), Augmentation=Cutout2018.0513.4—
Π-Model2022.0214.01—
Wide-ResNet-28-10 + AutoAugmentBackbone=Wide-ResNet-28-10, Augmentation=AutoAugment2018.0514.1—
AutoAugmentData Augmentation Strategy=AutoAugment2019.1114.1—
Faster AutoAugmentData Augmentation Strategy=Faster AutoAugment2019.1114.8—
IAM-Dlabeled samples=40002026.0515.07—
Pseudo-Labeling2022.0216.09—
Pseudo-LabelModel=WRN-28-2, # Param=1.5M2019.0416.21—
Wide-ResNet-28-10Backbone=Wide-ResNet-28-10, Augmentation=Cutout2018.0516.5—
Shake-Shake (26 2x96d)Backbone=Shake-Shake (26 2x96d), Augmentation=Baseline2018.0517.1—
Triple-GAN-V2Data augmentation=true, Classifier architecture=13-layer CNN, Number of runs=32019.1217.34—
MTData augmentation=true, Classifier architecture=13-layer CNN, Number of runs=32019.1218.1—
Wide-ResNet-28-10Backbone=Wide-ResNet-28-10, Augmentation=Baseline2018.0518.8—
SAMlabeled samples=40002026.0519.95—
Triple-GAN-V2Data augmentation=false, Classifier architecture=13-layer CNN, Number of runs=32019.1220.07—
Conv-Large, Γ-modelnumber of used labels=4000, data augmentation=false2015.0720.4—
MTData augmentation=false, Classifier architecture=13-layer CNN, Number of runs=32019.1221.02—
SGDlabeled samples=40002026.0522.45—
CutoutData Augmentation Strategy=Cutout2019.1122.5—
Conv-Large, supervised onlynumber of used labels=4000, data augmentation=false, supervised only=true2015.0723.33—
BaselineData Augmentation Strategy=None2019.1124.3—
Spike-and-Slab Sparse Codingnumber of used labels=4000, data augmentation=false2015.0731.9—
Triple-GAN-V1Data augmentation=false, Classifier architecture=13-layer CNN, Number of runs=32019.1232.73—
DADAData augmentation=false, Classifier architecture=13-layer CNN, Number of runs=32019.1233.57—
CNN (our code base)Data augmentation=false, Classifier architecture=13-layer CNN, Number of runs=32019.1246.1—
Baselinenetwork=WRN-28-22020.04—86.08
CLnetwork=WRN-28-22020.04—94.02
DAGnetwork=CNN-132020.04—93.87
DCTnetwork=CNN-132020.04—90.97
DMTnetwork=WRN-28-22020.04—94.21
DSnetwork=CNN-132020.04—91.11
EnAETEpochs=1024, Architecture=WideResNet-28-22021.04—94.7
FixMatchArchitecture=WideResNet-28-22021.04—95.7
ICTEpochs=600, Architecture=WideResNet-28-22021.04—92.7
LGA + VATArchitecture=WideResNet-28-22021.04—87.9
Mean TeacherEpochs=300, Architecture=WideResNet-28-22021.04—84.1
MixMatchArchitecture=WideResNet-28-22021.04—93.8
MixMatchnetwork=WRN-28-22020.04—93.76
MPLEpochs=2564, Architecture=WideResNet-28-22021.04—96.1
MTnetwork=WRN-28-22020.04—89.64
PAWS-NNEpochs=600, Architecture=WideResNet-28-2, Parameters=1.5M2021.04—96
ReMixMatchArchitecture=WideResNet-28-22021.04—94.9
SimCLRv2Architecture=ResNet-200 (+SK), Parameters=95M, Epochs=8002021.04—96
SimCLRv2Architecture=ResNet-18 (+SK), Parameters=12M, Epochs=8002021.04—92.1
Supervised Learning with full datasetEpochs=1000, Architecture=WideResNet-28-22021.04—94.9
Temporal EnsembleEpochs=300, Architecture=WideResNet-28-22021.04—83.6
UDAEpochs=2564, Architecture=WideResNet-28-22021.04—94.5
VAT + EntMinEpochs=123, Architecture=WideResNet-28-22021.04—86.9