Classification on CIFAR-10
98.8AccuracyDGMMC-S
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DGMMC-SEmbedding=ImageBind, G (Number of Gaussians)=22024.10 | 98.8 | — | — | |
| CoAt-CBMInterpretability=Yes2026.04 | 98.51 | — | — | |
| Linear ProbeInterpretability=No2026.04 | 97.93 | — | — | |
| HybridCBMInterpretability=Yes2026.04 | 97.91 | — | — | |
| ResCBMInterpretability=Yes2026.04 | 97.83 | — | — | |
| BYOLArchitecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Fine-tuned2020.06 | 97.8 | — | — | |
| DOT-CBMInterpretability=Yes2026.04 | 97.75 | — | — | |
| SimCLRArchitecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Fine-tuned2020.06 | 97.7 | — | — | |
| LaboInterpretability=Yes2026.04 | 97.7 | — | — | |
| Supervised-INArchitecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Fine-tuned2020.06 | 97.5 | — | — | |
| PCBM-hInterpretability=No2026.04 | 97.41 | — | — | |
| SimCLR (repro)Architecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Fine-tuned2020.06 | 97.4 | — | — | |
| F-Deep-ensBackbone=WideResNet28x102023.02 | 97.11 | 0.0851 | 0.59 | |
| F-MC-DropoutBackbone=WideResNet28x102023.02 | 97.1 | 0.0966 | 0.95 | |
| F-SWAGBackbone=WideResNet28x102023.02 | 97.09 | 0.0883 | 0.36 | |
| F-SWAG-DiagBackbone=WideResNet28x102023.02 | 97.05 | 0.0888 | 0.43 | |
| PCBMInterpretability=Yes2026.04 | 97.02 | — | — | |
| PEcoreData size=5B, Zero-shot=true2026.03 | 97 | — | — | |
| Deep-ensBackbone=WideResNet28x102023.02 | 96.96 | 0.1031 | 0.87 | |
| OpenCLIP ViT-H/14Zero-shot=true, Backbone=ViT-H/14, Confidence Interval=95% CI, Runs=52025.10 | 96.87 | — | — | |
| F-Deep-ensBackbone=PreResNet-1642023.02 | 96.7 | 0.1031 | 0.57 | |
| F-SGLDBackbone=WideResNet28x102023.02 | 96.43 | 0.1336 | 3.85 | |
| SWAG-DiagBackbone=WideResNet28x102023.02 | 96.41 | 0.1077 | 0.47 | |
| MC-DropoutBackbone=WideResNet28x102023.02 | 96.39 | 0.1094 | 0.94 | |
| F-MC-DropoutBackbone=PreResNet-1642023.02 | 96.39 | 0.1137 | 1.18 | |
| Deep-ensBackbone=PreResNet-1642023.02 | 96.39 | 0.1277 | 1.08 | |
| SWAGBackbone=WideResNet28x102023.02 | 96.32 | 0.1122 | 0.88 | |
| F-SWAGBackbone=PreResNet-1642023.02 | 96.25 | 0.1106 | 0.56 | |
| F-SWAG-DiagBackbone=PreResNet-1642023.02 | 96.23 | 0.1108 | 0.43 | |
| MC-DropoutBackbone=PreResNet-1642023.02 | 96.18 | 0.127 | 1.62 | |
| SWAG-DiagBackbone=PreResNet-1642023.02 | 96.03 | 0.1251 | 0.82 | |
| SWAGBackbone=PreResNet-1642023.02 | 96.03 | 0.1232 | 0.53 | |
| Random initArchitecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Fine-tuned2020.06 | 95.9 | — | — | |
| SGLDBackbone=WideResNet28x102023.02 | 95.87 | 0.1573 | 4.63 | |
| MetaCLIPData size=2.5B, Zero-shot=true2026.03 | 95.7 | — | — | |
| LlipData size=2.5B, Zero-shot=true2026.03 | 95.7 | — | — | |
| F-SGLDBackbone=PreResNet-1642023.02 | 95.04 | 0.1912 | 6.01 | |
| OpenCLIPData size=2B, Zero-shot=true2026.03 | 94.8 | — | — | |
| SigLIP2Data size=10B, Zero-shot=true2026.03 | 94.8 | — | — | |
| SGLDBackbone=PreResNet-1642023.02 | 94.79 | 0.2089 | 7.11 | |
| BayesVLMBackbone=OpenCLIP ViT-B-32, Mode=Zero-shot2024.12 | 93.62 | 0.2 | 0.72 | |
| CLIPBackbone=OpenCLIP ViT-B-32, Mode=Zero-shot2024.12 | 93.61 | 0.21 | 1.45 | |
| CLIP (temp. scaling)Backbone=OpenCLIP ViT-B-32, Mode=Zero-shot, Calibration=Temperature Scaling2024.12 | 93.61 | 0.21 | 1.88 | |
| Supervised-INArchitecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Linear evaluation2020.06 | 93.6 | — | — | |
| PLIFMethod=Spike-based BP2022.06 | 93.5 | — | — | |
| COSMOSData size=30M, Zero-shot=true2026.03 | 93.4 | — | — | |
| STBP-tdBNMethod=ANN2SNN2022.06 | 93.16 | — | — | |
| FLAIRData size=30M, Zero-shot=true2026.03 | 93.1 | — | — | |
| GoldiCLIPData size=30M, Zero-shot=true2026.03 | 92.9 | — | — | |
| STL-SNNMethod=Spike-based BP, Selection=Best top-12022.06 | 92.42 | — | — | |
| DreamLIPData size=30M, Zero-shot=true2026.03 | 92.3 | — | — | |
| SigLIPData size=10B, Zero-shot=true2026.03 | 92.3 | — | — | |
| CLIPData size=30M, Zero-shot=true2026.03 | 92.2 | — | — | |
| CLIPBackbone=CLIP ViT-L-14, zero-shot=true2023.05 | 91.7 | — | — | |
| POS PCABackbone=CLIP ViT-L-14, zero-shot=true2023.05 | 91.7 | — | — | |
| POS PGABackbone=CLIP ViT-L-14, zero-shot=true2023.05 | 91.6 | — | — | |
| Spike-NormMethod=ANN2SNN2022.06 | 91.55 | — | — | |
| TSSL-BPMethod=Spike-based BP2022.06 | 91.41 | — | — | |
| BYOLArchitecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Linear evaluation2020.06 | 91.3 | — | — | |
| SigLIPData size=30M, Zero-shot=true2026.03 | 91 | — | — | |
| SimCLRArchitecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Linear evaluation2020.06 | 90.6 | — | — | |
| Noun SubmanifoldBackbone=CLIP ViT-L-14, zero-shot=true2023.05 | 90.6 | — | — | |
| NeuNormMethod=Spike-based BP2022.06 | 90.53 | — | — | |
| SimCLR (repro)Architecture=ResNet-50, Pre-training Dataset=ImageNet, Evaluation Protocol=Linear evaluation2020.06 | 90.5 | — | — | |
| Partial G-CNNsBase Group=E(2), No. Elems=162021.10 | 90.12 | — | — | |
| Venkatesha+Hardware=Simulated, Number of nodes=5+2026.03 | 90 | — | — | |
| SDGM-FEvaluation Protocol=end-to-end pipeline2024.10 | 89.54 | — | — | |
| Partial G-CNNsBase Group=E(2), No. Elems=82021.10 | 89 | — | — | |
| Partial G-CNNsBase Group=SE(2), No. Elems=82021.10 | 88.59 | — | — | |
| Partial G-CNNsBase Group=SE(2), No. Elems=162021.10 | 88.59 | — | — | |
| TTABackbone=OpenCLIP ViT-B-32, Mode=Zero-shot, Augmentation=Test-time Augmentation2024.12 | 88.54 | 0.35 | 2.02 | |
| SD, trunc. inverse-SNRZero-shot=true, Backbone=Stable Diffusion, Weighting=truncated inverse-SNR, Confidence Interval=95% CI, Runs=52025.10 | 88.38 | — | — | |
| SD, inverse-SNRZero-shot=true, Backbone=Stable Diffusion, Weighting=inverse-SNR, Confidence Interval=95% CI, Runs=52025.10 | 87.26 | — | — | |
| Label-FreeInterpretability=Yes2026.04 | 86.42 | — | — | |
| SD, uniform (ours)Zero-shot=true, Backbone=Stable Diffusion, Weighting=uniform, Confidence Interval=95% CI, Runs=52025.10 | 85.17 | — | — | |
| SD, uniform (Li et al.)Zero-shot=true, Backbone=Stable Diffusion, Weighting=uniform, Confidence Interval=95% CI, Runs=52025.10 | 84.67 | — | — | |
| AugerinoBase Group=E(2), No. Elems=82021.10 | 84.29 | — | — | |
| AugerinoBase Group=E(2), No. Elems=162021.10 | 83.54 | — | — | |
| xCLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.10 | 83.4 | — | — | |
| TandomSNNMethod=ANN2SNN2022.06 | 82.78 | — | — | |
| AugerinoBase Group=SE(2), No. Elems=162021.10 | 82.53 | — | — | |
| AugerinoBase Group=SE(2), No. Elems=82021.10 | 82.38 | — | — | |
| CLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.10 | 79.7 | — | — | |
| nCLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.10 | 79.5 | — | — | |
| CLIP RN-50Zero-shot=true, Backbone=RN-50, Confidence Interval=95% CI, Runs=52025.10 | 75.42 | — | — | |
| STBPMethod=Spike-based BP2022.06 | 50.7 | — | — | |
| APFLprotocol=Known Association2025.11 | 44.7 | — | — | |
| pFedMeprotocol=Known Association2025.11 | 42.4 | — | — | |
| FLUX-priorprotocol=Known Association2025.11 | 40.1 | — | — | |
| FLUX-priorprotocol=Test Phase2025.11 | 39.3 | — | — | |
| FLUXprotocol=Test Phase2025.11 | 38.7 | — | — | |
| FLUXprotocol=Known Association2025.11 | 38.6 | — | — | |
| IFCAprotocol=Known Association2025.11 | 38.3 | — | — | |
| IFCAprotocol=Test Phase2025.11 | 36.6 | — | — | |
| APFLprotocol=Test Phase2025.11 | 36.6 | — | — | |
| FeSEMprotocol=Known Association2025.11 | 36.5 | — | — | |
| FeSEMprotocol=Test Phase2025.11 | 35.3 | — | — | |
| FedDriftprotocol=Known Association2025.11 | 35.1 | — | — | |
| ATPprotocol=Test Phase2025.11 | 33.6 | — | — | |
| CFLprotocol=Test Phase2025.11 | 33.2 | — | — |