Classification on MNIST (Accuracy)
99.3AccuracyDGMMC-S
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DGMMC-SEmbedding=ImageBind, G (Number of Gaussians)=12024.10 | 99.3 | — | |
| SDGM-FEvaluation Protocol=end-to-end pipeline2024.10 | 99.14 | — | |
| Multilinear-CovJac2026.05 | 98.36 | — | |
| Soft-Mix2026.05 | 98.32 | — | |
| M-CovJac#p/n=4, L=6, k=64k, iterations=50k, τ=1.02026.05 | 98.3 | — | |
| Soft-Mix#p/n=16, L=6, k=64k, iterations=50k, τ=1.02026.05 | 98.29 | — | |
| Gumbel-ST2026.05 | 98.21 | — | |
| Gumbel-ST#p/n=16, L=6, k=64k, iterations=50k, τ=1.02026.05 | 98.17 | — | |
| Multilinear-STE2026.05 | 98.15 | — | |
| M-STE#p/n=4, L=6, k=64k, iterations=50k, τ=1.02026.05 | 98.09 | — | |
| Bullet TrainsModel Architecture=1F350H, Exact Gradients=true, Continuous Spike Times=true, Parallelized=true, Compute Depth=O(C log K), Memory=O(N)2026.02 | 98.04 | — | |
| DNN (NhL = 2)Number of hidden units=10582025.01 | 98.02 | — | |
| 1F350H (Wunderlich & Pehle)Exact Gradients=true, Continuous Spike Times=true, Parallelized=false, Compute Depth=O(N), Memory=O(N)2026.02 | 97.6 | — | |
| HaSiSTArchitecture=784–200–10, Coding scheme=rate-based, Multiplication-Free=false2026.04 | 97.5 | — | |
| S4NNArchitecture=784–400–10, Coding scheme=Single-Spike, Multiplication-Free=false2026.04 | 97.4 | — | |
| STiDi-BPArchitecture=784–400–10, Coding scheme=Single-Spike, Multiplication-Free=false2026.04 | 97.4 | — | |
| Multiplication-Free Spike-Time Learning AlgorithmArchitecture=784–400–10, Coding scheme=Single-Spike, Multiplication-Free=true2026.04 | 97.4 | — | |
| 1F350H, τm = 2τsExact Gradients=true, Continuous Spike Times=true, Parallelized=false, Compute Depth=O(N), Memory=O(N)2026.02 | 97.2 | — | |
| MostafaArchitecture=784–800–10, Coding scheme=Single-Spike, Multiplication-Free=false2026.04 | 97.2 | — | |
| BS4NNArchitecture=784–600–10, Coding scheme=Single-Spike, Multiplication-Free=false2026.04 | 97 | — | |
| TabNetFeature extraction=ResNet50, Subsampling=11K2026.05 | 96.88 | — | |
| STiDi-BPArchitecture=784–400–400–10, Coding scheme=Single-Spike, Multiplication-Free=false2026.04 | 96.8 | — | |
| HaSiSTArchitecture=784–400–400–10, Coding scheme=rate-based, Multiplication-Free=false2026.04 | 96.8 | — | |
| LSPINFeature extraction=ResNet50, Subsampling=11K2026.05 | 96.77 | — | |
| S4NNArchitecture=784–400–400–10, Coding scheme=Single-Spike, Multiplication-Free=false2026.04 | 96.7 | — | |
| DynaTabFeature extraction=ResNet50, Subsampling=11K2026.05 | 96.68 | — | |
| Original PNNNumber of hidden units=600002025.01 | 96.5 | — | |
| BS4NNArchitecture=784–400–400–10, Coding scheme=Single-Spike, Multiplication-Free=false2026.04 | 96.5 | — | |
| MostafaArchitecture=784–400–400–10, Coding scheme=Single-Spike, Multiplication-Free=false2026.04 | 96.5 | — | |
| Multiplication-Free Spike-Time Learning AlgorithmArchitecture=784–400–400–10, Coding scheme=Single-Spike, Multiplication-Free=true2026.04 | 96.5 | — | |
| VNNs Hyperplanes# Parameters=52.4K2026.05 | 96.4 | — | |
| LLSPINFeature extraction=ResNet50, Subsampling=11K2026.05 | 96.36 | — | |
| MLPFeature extraction=ResNet50, Subsampling=11K2026.05 | 96.2 | — | |
| MLP# Parameters=238.3K2026.05 | 96 | — | |
| tSNEClassifier=Extra Trees (ET), Gate layer=Disabled (λ=0)2022.07 | 95.9 | — | |
| UDRNClassifier=Extra Trees (ET), Gate layer=Disabled (λ=0)2022.07 | 95.9 | — | |
| FLUX-priorprotocol=Known Association2025.11 | 95.8 | — | |
| FLUX-priorprotocol=Test Phase2025.11 | 95.7 | — | |
| pFedMeprotocol=Known Association2025.11 | 95.5 | — | |
| APFLprotocol=Known Association2025.11 | 95.5 | — | |
| LGBMFeature extraction=ResNet50, Subsampling=11K2026.05 | 95.42 | — | |
| CS-PNNNumber of hidden units=36842025.01 | 94.9 | — | |
| UMAPClassifier=Extra Trees (ET), Gate layer=Disabled (λ=0)2022.07 | 94.5 | — | |
| UDRN2022.07 | 94.3 | — | |
| PUMAPClassifier=Extra Trees (ET), Gate layer=Disabled (λ=0)2022.07 | 94.2 | — | |
| FLUXprotocol=Test Phase2025.11 | 94 | — | |
| QS2022.07 | 93.2 | — | |
| FLUXprotocol=Known Association2025.11 | 92.5 | — | |
| CAE2022.07 | 92.1 | — | |
| FedDriftprotocol=Known Association2025.11 | 91.1 | — | |
| NDFS2022.07 | 90.4 | — | |
| FeSEMprotocol=Known Association2025.11 | 87.8 | — | |
| AEFS2022.07 | 86.4 | — | |
| CFLprotocol=Test Phase2025.11 | 86.1 | — | |
| FedAvgprotocol=Test Phase2025.11 | 85.6 | — | |
| ATPprotocol=Test Phase2025.11 | 85.6 | — | |
| APFLprotocol=Test Phase2025.11 | 84.7 | — | |
| IFCAprotocol=Known Association2025.11 | 84.1 | — | |
| FeSEMprotocol=Test Phase2025.11 | 82.8 | — | |
| CFLprotocol=Known Association2025.11 | 81.3 | — | |
| FedAvgprotocol=Known Association2025.11 | 80.9 | — | |
| MVT + FTArch=ViT-L2025.12 | 79.2 | — | |
| IFCAprotocol=Test Phase2025.11 | 78.2 | — | |
| Vanilla FTArch=ViT-L2025.12 | 77.5 | — | |
| GRAEClassifier=Extra Trees (ET), Gate layer=Disabled (λ=0)2022.07 | 77.2 | — | |
| CLIPArch=ViT-L2025.12 | 76.4 | — | |
| MCFS2022.07 | 76 | — | |
| FAE2022.07 | 70.5 | — | |
| IVISClassifier=Extra Trees (ET), Gate layer=Disabled (λ=0)2022.07 | 68.3 | — | |
| MVT + FTArch=ViT-g2025.12 | 65.7 | — | |
| FedDriftprotocol=Test Phase2025.11 | 65.1 | — | |
| MVTArch=ViT-g2025.12 | 63.2 | — | |
| Vanilla FTArch=ViT-g2025.12 | 62.9 | — | |
| EVAArch=ViT-g2025.12 | 62.2 | — | |
| CLIPArch=RN502025.12 | 58.5 | — | |
| VQAArch=ViT-L2025.12 | 56.7 | — | |
| VQAArch=ViT-g2025.12 | 55.7 | — | |
| MVTArch=ViT-L2025.12 | 53 | — | |
| CLIPArch=RN1012025.12 | 51.6 | — | |
| FedEMprotocol=Test Phase2025.11 | 50 | — | |
| FedRCprotocol=Test Phase2025.11 | 49.9 | — | |
| FedEMprotocol=Known Association2025.11 | 48.3 | — | |
| CLIPArch=ViT-B2025.12 | 47.9 | — | |
| FedRCprotocol=Known Association2025.11 | 47.5 | — | |
| IVFS2022.07 | 42.4 | — | |
| LS2022.07 | 17 | — | |
| xCLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.10 | 12.3 | — | |
| CLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.10 | 10.5 | — | |
| nCLIPBackbone=ViT-B/16, Pre-trained on=IT35M, Mode=Zero-shot2022.10 | 9.9 | — |