List your company
ResearchFormerly WizWand

Image Classification on SVHN (Standard Accuracy)

98.6AccuracyNoisy Student Training
4.126428.653253.1877.7068Nov 11, 2019Dec 10, 2020Jan 9, 2022Feb 8, 2023Mar 9, 2024Apr 8, 2025May 9, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Noisy Student TrainingBackbone=B02019.1198.6
RandAugmentBackbone=WRN2019.1198.3
RoPEPositional Encoding=RoPE2026.0398.15
PaRaMSModel Status=Protected Standalone (θ̂_def)2025.1198.11
RandAugmentBackbone=EfficientNet-B02019.1198.1
No PEPositional Encoding=None2026.0398.09
Learnable PEPositional Encoding=Learnable2026.0398.09
ResNet-18Backbone=ResNet-182026.0397.98
AdapterTuneBackbone=ViT-B/16, Trainable Parameters (%)=0.9%2026.0397.5
Full fine-tuningBackbone=ViT-B/16, Trainable Parameters (%)=100%2026.0397.5
Full fine-tuningBackbone=ViT-S/16, Trainable Parameters (%)=100%2026.0397.4
Full fine-tuningBackbone=DeiT-T, Trainable Parameters (%)=100%2026.0397.2
MergeGuardModel Status=Protected Standalone (θ̂_def)2025.1196.82
Fine-TunedBackbone=CLIP ViT-B/32, Classes=102025.1296.4
AdapterTuneBackbone=ViT-S/16, Trainable Parameters (%)=0.9%2026.0396.2
AdapterTuneBackbone=DeiT-T, Trainable Parameters (%)=0.9%2026.0395.3
ADAProtection Status=None (theta_merge)2025.1193.4
TIESProtection Status=None (theta_merge)2025.1190.3
M-TIESBackbone=ViT-L/142026.0289.78
TIESBackbone=ViT-L/142026.0289.42
DAREBackbone=ViT-L/142026.0289.19
TAProtection Status=None (theta_merge)2025.1187.9
TIESBackbone=ViT-B/322026.0286.2
Task ArithmeticBackbone=ViT-L/142026.0285.26
EnsemblingBackbone=ViT-L/142026.0284.92
DAREBackbone=ViT-B/322026.0283.96
M-TIESBackbone=ViT-B/322026.0283.06
EnsemblingBackbone=ViT-B/322026.0282.15
Simple Avg.Backbone=ViT-L/142026.0278.23
WAProtection Status=None (theta_merge)2025.1178.2
Task ArithmeticBackbone=ViT-B/322026.0276.68
PaRaMSModel Status=Merged (θ̂_merge)2025.1171.95
M-CovJac#p/n=4, L=6, k=128k, iterations=50k, τ=1.02026.0568.91
Soft-Mix#p/n=16, L=6, k=128k, iterations=50k, τ=1.02026.0568.21
Task MatrixBackbone=CLIP ViT-B/32, Best Layer=8, Classes=102025.1266.7
M-STE#p/n=4, L=6, k=128k, iterations=50k, τ=1.02026.0566.69
Linear ProbeBackbone=CLIP ViT-B/32, Classes=102025.1266.6
Gumbel-ST#p/n=16, L=6, k=128k, iterations=50k, τ=1.02026.0566.12
Head-only tuningBackbone=ViT-B/16, Trainable Parameters (%)=0.1%2026.0365.5
Simple Avg.Backbone=ViT-B/322026.0264.16
Head-only tuningBackbone=ViT-S/16, Trainable Parameters (%)=0.1%2026.0354.5
TAProtection Status=With MergeGuard (theta_hat_merge)2025.1147.28
MergeGuardModel Status=Merged (θ̂_merge)2025.1147.28
Head-only tuningBackbone=DeiT-T, Trainable Parameters (%)=0.1%2026.0344.5
WAProtection Status=With MergeGuard (theta_hat_merge)2025.1110.05
TIESProtection Status=With MergeGuard (theta_hat_merge)2025.117.77
ADAProtection Status=With MergeGuard (theta_hat_merge)2025.117.76