Image Classification on Fashion MNIST
98.8AccuracyOPAL
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| OPALManual labelling effort=Identical to RQ1, Fine-tuning=85%2025.07 | 98.8 | — | — | |
| SOTA2024.06 | 96.91 | — | — | |
| Fine-Tuning DARTSData Augmentation=cutout + random erasing, Params(M)=3.2, Search Method=Gradient-Based2020.06 | 96.91 | — | — | |
| DARTS(2nd order)Data Augmentation=cutout + random erasing, Params(M)=2.6, Search Method=Gradient-Based2020.06 | 96.57 | — | — | |
| FMixModel=ResNet2020.02 | 96.36 | — | — | |
| MixUpModel=Dense2020.02 | 96.3 | — | — | |
| MixUpModel=ResNet2020.02 | 96.28 | — | — | |
| FMixModel=Dense2020.02 | 96.26 | — | — | |
| CutMixModel=Dense2020.02 | 96.12 | — | — | |
| CutMixModel=ResNet2020.02 | 96.03 | — | — | |
| FMixModel=WRN2020.02 | 96 | — | — | |
| BaselineModel=Dense2020.02 | 95.84 | — | — | |
| MixUpModel=WRN2020.02 | 95.75 | — | — | |
| BaselineModel=ResNet2020.02 | 95.7 | — | — | |
| CutMixModel=WRN2020.02 | 95.64 | — | — | |
| B-VGGManual labelling effort=Identical to RQ1, Fine-tuning=85%, Backbone=VGG2025.07 | 95.6 | — | — | |
| VGG8BParams(M)=7.3, Search Method=Manual2020.06 | 95.47 | — | — | |
| B-PseudoManual labelling effort=Identical to RQ1, Fine-tuning=85%2025.07 | 95.3 | — | — | |
| BaselineModel=WRN2020.02 | 95.29 | — | — | |
| OursNumber of Parameters=1,028,2342022.07 | 95.03 | — | — | |
| B-ResNetManual labelling effort=Identical to RQ1, Fine-tuning=85%, Backbone=ResNet2025.07 | 95 | — | — | |
| STL-SNNMethod=Spike-based BP, Selection=Best top-12022.06 | 94.47 | — | — | |
| DeepCapsParams(M)=7.2, Search Method=Manual2020.06 | 94.46 | — | — | |
| Inception-V3Number of Parameters=23,851,7842022.07 | 94.44 | — | — | |
| PLIFMethod=Spike-based BP2022.06 | 94.38 | — | — | |
| B-ViTManual labelling effort=Identical to RQ1, Fine-tuning=85%, Backbone=ViT2025.07 | 93.7 | — | — | |
| OnDev-LCT-2/3#Params=0.35M, MACs=0.08G2024.01 | 93.31 | — | — | |
| IndividualBackbone=ViT-B/16, Pre-trained=ImageNet-21k, Number of Merged Models=12024.05 | 93.26 | — | — | |
| R-ExplaiNet26-64Parameter Size=0.89MP2024.10 | 93.03 | — | — | |
| TSSL-BPMethod=Spike-based BP2022.06 | 92.83 | — | — | |
| ResNet26-64Parameter Size=0.89MP2024.10 | 92.83 | — | — | |
| ResNet-32#Params=0.47M, MACs=0.07G2024.01 | 92.82 | — | — | |
| OnDev-LCT-8/3#Params=0.95M, MACs=0.12G2024.01 | 92.82 | — | — | |
| OnDev-LCT-1/1#Params=0.21M, MACs=0.03G2024.01 | 92.73 | — | — | |
| OnDev-LCT-1/3#Params=0.25M, MACs=0.07G2024.01 | 92.72 | — | — | |
| OnDev-LCT-4/1#Params=0.51M, MACs=0.05G2024.01 | 92.7 | — | — | |
| OnDev-LCT-4/3#Params=0.55M, MACs=0.09G2024.01 | 92.6 | — | — | |
| OnDev-LCT-8/1#Params=0.91M, MACs=0.08G2024.01 | 92.59 | — | — | |
| OnDev-LCT-2/1#Params=0.31M, MACs=0.04G2024.01 | 92.5 | — | — | |
| ResNet-56#Params=0.86M, MACs=0.13G2024.01 | 92.48 | — | — | |
| ResNet-44#Params=0.67M, MACs=0.10G2024.01 | 92.46 | — | — | |
| ResNet-20#Params=0.27M, MACs=0.04G2024.01 | 92.45 | — | — | |
| EMR-MERGINGBackbone=ViT-B/16, Pre-trained=ImageNet-21k, Number of Merged Models=302024.05 | 92.41 | — | — | |
| NeupdeParams(M)=0.4, Search Method=Manual2020.06 | 92.4 | — | — | |
| S2-STDP+NCGNeurons per class=5, Feature Extractor=SoftHebb-CNN2024.10 | 91.86 | — | — | |
| CCT-4/2#Params=0.48M, MACs=0.05G2024.01 | 91.73 | — | — | |
| IDGPBackbone=LeNet-52022.04 | 91.48 | — | — | |
| CCT-2/2#Params=0.28M, MACs=0.03G2024.01 | 91.37 | — | — | |
| Wilson-Cowan model for metapopulationBackbone=simple CNN2024.06 | 91.35 | 0.0016 | — | |
| SSTDP+NCGNeurons per class=5, Feature Extractor=SoftHebb-CNN2024.10 | 91.06 | — | — | |
| ViT-Lite-2/4#Params=1.23M, MACs=0.07G2024.01 | 90.7 | — | — | |
| VALENBackbone=LeNet-52022.04 | 90.63 | — | — | |
| S2-STDPNeurons per class=1, Feature Extractor=SoftHebb-CNN2024.10 | 90.61 | — | — | |
| BPTTNetwork structure=400 (R400), Time steps=52022.10 | 90.58 | — | — | |
| MobileNetv2/1.0#Params=2.27M, MACs=0.01G2024.01 | 90.53 | — | — | |
| OTTTONetwork structure=400 (R400), Time steps=52022.10 | 90.4 | — | — | |
| OTTTANetwork structure=400 (R400), Time steps=52022.10 | 90.36 | — | — | |
| ViT-Lite-1/4#Params=1.21M, MACs=0.04G2024.01 | 90.25 | — | — | |
| Total Variation divergenceNoise=Random (0.2)2020.11 | 90.22 | — | — | |
| CNNEnc. type=-, Format=-2026.03 | 90.2 | — | — | |
| MobileNetv2/0.2#Params=0.21M, MACs=<0.01G2024.01 | 90.16 | — | — | |
| ST-RSBPMethod=Spike-based BP2022.06 | 90.13 | — | — | |
| MobileNetv2/0.75#Params=1.39M, MACs=<0.01G2024.01 | 90.13 | — | — | |
| PLCRBackbone=LeNet-52022.04 | 90.1 | — | — | |
| XGBoostEnc. type=-, Format=-2026.03 | 90.1 | — | — | |
| AT + IGRAttack Protocol=Natural2022.05 | 89.98 | — | — | |
| CCBackbone=LeNet-52022.04 | 89.86 | — | — | |
| PRODENBackbone=LeNet-52022.04 | 89.79 | — | — | |
| Peer LossNoise=Random (0.2)2020.11 | 89.78 | — | — | |
| Kullback-Leibler divergenceNoise=Sparse, Low2020.11 | 89.77 | — | — | |
| ATAttack Protocol=Natural2022.05 | 89.75 | — | — | |
| Total Variation divergenceNoise=Sparse, Low2020.11 | 89.74 | — | — | |
| Jensen-Shannon divergenceNoise=Random (0.2)2020.11 | 89.73 | — | — | |
| MobileNetv2/0.5#Params=0.72M, MACs=<0.01G2024.01 | 89.7 | — | — | |
| RCBackbone=LeNet-52022.04 | 89.6 | — | — | |
| SSTDPNeurons per class=1, Feature Extractor=SoftHebb-CNN2024.10 | 89.36 | — | — | |
| Peer LossNoise=Uniform, Low2020.11 | 89.31 | — | — | |
| Kullback-Leibler divergenceNoise=Random (0.2)2020.11 | 89.24 | — | — | |
| Total Variation divergenceNoise=Uniform, Low2020.11 | 89 | — | — | |
| Linear C-SVMEnc. type=-, Format=-2026.03 | 89 | — | — | |
| LWSBackbone=LeNet-52022.04 | 88.99 | — | — | |
| Jensen-Shannon divergenceNoise=Sparse, Low2020.11 | 88.8 | — | — | |
| S2-STDP+NCGNeurons per class=5, Feature Extractor=STDP-CSNN2024.10 | 88.72 | — | — | |
| Laplace-HDCEnc. type=CB, Format=BP/FP322026.03 | 88.7 | — | — | |
| Jensen-Shannon divergenceNoise=Uniform, Low2020.11 | 88.58 | — | — | |
| Binary MM-HDC (OvO)Enc. type=NL, Format=FP322026.03 | 88.5 | — | — | |
| PICOBackbone=LeNet-52022.04 | 88.41 | — | — | |
| DMI lossNoise=Sparse, Low2020.11 | 88.32 | — | — | |
| Kullback-Leibler divergenceNoise=Uniform, Low2020.11 | 88.32 | — | — | |
| Peer LossNoise=Sparse, Low2020.11 | 88.15 | — | — | |
| MLPEnc. type=-, Format=-2026.03 | 88.1 | — | — | |
| R-STDPNeurons per class=20, Feature Extractor=SoftHebb-CNN2024.10 | 88.06 | — | — | |
| no-stitchstitching_type=none, decoder=SVM (linear kernel)2023.11 | 88 | — | — | |
| FedAvgAttack Type=Gradient Manipulation, Proportion of Attackers=0%2024.05 | 88 | — | — | |
| FedAvgAttack Type=Label Flipping, Proportion of Attackers=0%2024.05 | 88 | — | — | |
| CAVLBackbone=LeNet-52022.04 | 87.94 | — | — | |
| Jensen-Shannon divergenceNoise=Random (0.7)2020.11 | 87.79 | — | — | |
| ViT-Lite-2/16#Params=0.27M, MACs=<0.01G2024.01 | 87.67 | — | — | |
| SSTDP+NCGNeurons per class=5, Feature Extractor=STDP-CSNN2024.10 | 87.59 | — | — | |
| RFF-HDCEnc. type=CB, Format=BP/FP322026.03 | 87.4 | — | — |