Image Classification on Fashion-MNIST (train)
100Accuracy (Train)Vanilla
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| VanillaOptimizer=Adam, Nlinear in [%]=100, Nconv in [%]=1002021.05 | 100 | — | — | — | — | — | |
| VanillaOptimizer=Adam2021.05 | 100 | 100 | — | — | — | — | |
| GLassoOptimizer=SGD + thresh., Nlinear in [%]=3.6, Nconv in [%]=4.32021.05 | 94.8 | — | — | — | — | — | |
| LassoOptimizer=SGD + thresh.2021.05 | 94.7 | 3.5 | — | — | — | — | |
| LinBreg with MomentumStrategy=Bregman, Optimizer=LinBreg (β = 0.9)2021.05 | 93.8 | 2.7 | — | — | — | — | |
| AdaBregStrategy=Bregman, Optimizer=AdaBreg2021.05 | 93.6 | 2.3 | — | — | — | — | |
| LinBreg (β = 0.9)Strategy=Bregman, Optimizer=LinBreg (β = 0.9), Nlinear in [%]=3.5, Nconv in [%]=4.72021.05 | 93.5 | — | — | — | — | — | |
| LinBregStrategy=Bregman, Optimizer=LinBreg, Nlinear in [%]=3.8, Nconv in [%]=4.22021.05 | 93.1 | — | — | — | — | — | |
| MLPHyperpars=h = 64, Params=148,5862026.06 | 93.08 | — | — | — | — | — | |
| AdaBregStrategy=Bregman, Optimizer=AdaBreg, Nlinear in [%]=3.5, Nconv in [%]=2.82021.05 | 92.6 | — | — | — | — | — | |
| Pruning (5%)Optimizer=SGD2021.05 | 92 | 4.7 | — | — | — | — | |
| GLassoOptimizer=ProxGD, Nlinear in [%]=3.0, Nconv in [%]=3.72021.05 | 91.6 | — | — | — | — | — | |
| LassoOptimizer=ProxGD2021.05 | 91.4 | 4.8 | — | — | — | — | |
| LPParams=23,4662026.06 | 91.39 | — | — | — | — | — | |
| HalfNetHyperpars=k = 8, Params=21,9302026.06 | 91.22 | — | — | — | — | — | |
| MLPHyperpars=h = 20, Params=46,7262026.06 | 91.12 | — | — | — | — | — | |
| LinBregStrategy=Bregman, Optimizer=LinBreg2021.05 | 91.1 | 1.9 | — | — | — | — | |
| HalfNetHyperpars=k = 4, Params=12,7142026.06 | 90 | — | — | — | — | — | |
| Pruning (7%)Optimizer=SGD, Nlinear in [%]=7.0, Nconv in [%]=6.52021.05 | 89.9 | — | — | — | — | — | |
| Full model baselineModel=full, Params=55,050, Architecture=784–64–64–102026.06 | 89.75 | — | — | — | — | — | |
| MLPHyperparameters=h = 64, Number of Parameters=50,890, Architecture=784–64–102026.06 | 89.47 | — | — | — | — | — | |
| HalfNetModel=k = 48, Params=42,458, Architecture=784–64–64–102026.06 | 89.33 | — | — | — | — | — | |
| MLPHyperpars=h = 10, Params=23,5762026.06 | 89.1 | — | — | — | — | — | |
| HalfNetHyperpars=k = 2, Params=8,1062026.06 | 88.22 | — | — | — | — | — | |
| HalfNetModel=k = 32, Params=28,890, Architecture=784–64–64–102026.06 | 88.22 | — | — | — | — | — | |
| HalfNetHyperparameters=k = 16, Number of Parameters=14,106, Architecture=784–64–102026.06 | 87.96 | — | — | — | — | — | |
| HalfNetHyperpars=k = 1, Params=5,8022026.06 | 87.76 | — | — | — | — | — | |
| MLPHyperparameters=h = 20, Number of Parameters=15,910, Architecture=784–64–102026.06 | 87.29 | — | — | — | — | — | |
| HalfNetHyperparameters=k = 8, Number of Parameters=7,834, Architecture=784–64–102026.06 | 87 | — | — | — | — | — | |
| HalfNetModel=k = 16, Params=15,322, Architecture=784–64–64–102026.06 | 86.99 | — | — | — | — | — | |
| LPNumber of Parameters=7,850, Architecture=784–64–102026.06 | 86.35 | — | — | — | — | — | |
| HalfNetModel=k = 8, Params=8,538, Architecture=784–64–64–102026.06 | 86.2 | — | — | — | — | — | |
| HalfNetHyperparameters=k = 4, Number of Parameters=4,695, Architecture=784–64–102026.06 | 85.53 | — | — | — | — | — | |
| MLPHyperparameters=h = 10, Number of Parameters=7,910, Architecture=784–64–102026.06 | 85.2 | — | — | — | — | — | |
| HalfNetk=16, Params=14,1062026.06 | 85.13 | — | — | — | — | — | |
| HalfNetModel=k = 4, Params=5,146, Architecture=784–64–64–102026.06 | 84.43 | — | — | — | — | — | |
| HalfNetHyperparameters=k = 2, Number of Parameters=3,130, Architecture=784–64–102026.06 | 84.27 | — | — | — | — | — | |
| HalfNetHyperparameters=k = 1, Number of Parameters=2,346, Architecture=784–64–102026.06 | 83.49 | — | — | — | — | — | |
| Adaptive CFF (Learnable)Algorithm=Adaptive CFF (Learnable)2025.12 | 83.3 | — | — | — | — | — | |
| HalfNetModel=k = 2, Params=3,450, Architecture=784–64–64–102026.06 | 83.15 | — | — | — | — | — | |
| MLPHyperpars=random, Params=1,0662026.06 | 82.85 | — | — | — | — | — | |
| Collaborative FF (Fixed)Algorithm=Collaborative FF (Fixed)2025.12 | 82.8 | — | — | — | — | — | |
| Original FFAlgorithm=Original FF2025.12 | 82.4 | — | — | — | — | — | |
| HalfNetk=8, Params=7,8342026.06 | 82.4 | — | — | — | — | — | |
| HalfNetk=4, Params=4,6982026.06 | 81.69 | — | — | — | — | — | |
| HalfNetk=2, Params=3,1302026.06 | 77.16 | — | — | — | — | — | |
| randomParams=6502026.06 | 76.73 | — | — | — | — | — | |
| randomNumber of Parameters=650, Architecture=784–64–102026.06 | 76.6 | — | — | — | — | — | |
| Random baselineModel=random, Params=650, Architecture=784–64–64–102026.06 | 72.16 | — | — | — | — | — | |
| 0/1(BCD)Filter number=3800, Parameter number=4294k, FLOPs (M)=0.3632022.06 | — | — | 7.78 | — | — | — | |
| Bi-Real-NetFilter number=4000, Parameter number=5172k, FLOPs (M)=0.3822022.06 | — | — | 7.632 | — | — | — | |
| BPArchitecture=4-conv2026.06 | — | — | 3.09 | — | — | — | |
| CCNNFilter number=4000, Parameter number=5172k, FLOPs (M)=0.3822022.06 | — | — | 8.344 | — | — | — | |
| cEPArchitecture=4-conv2026.06 | — | — | 4.01 | — | — | — | |
| CNNSetting=Clean Training, Mode=clean2025.10 | — | — | — | — | 9.2 | — | |
| CNN (Clean)Model=CNN (Clean)2025.10 | — | — | 9.2 | — | — | — | |
| cSB-EPArchitecture=4-conv, Algorithm Variant=cEP2026.06 | — | — | 3.41 | — | — | — | |
| DNNSetting=Clean Training, Mode=clean2025.10 | — | — | — | — | 9 | — | |
| DNN (Clean)Model=DNN (Clean)2025.10 | — | — | 9 | — | — | — | |
| Evolved (APR)LR=Fixed, Optimizer=Adam, Epochs=1002026.03 | — | — | — | — | — | 0.0214 | |
| Evolved (APR)LR=Diminishing, Optimizer=Adam, Epochs=1002026.03 | — | — | — | — | — | 0.0222 | |
| Incremental Gradient (IG)LR=Fixed, Optimizer=Adam, Epochs=1002026.03 | — | — | — | — | — | 0.0388 | |
| Incremental Gradient (IG)LR=Diminishing, Optimizer=Adam, Epochs=1002026.03 | — | — | — | — | — | 0.0463 | |
| KPCA2026.03 | — | — | 30.5 | 35.173 | — | — | |
| KSIR2026.03 | — | — | 47 | 8.474 | — | — | |
| Original2026.03 | — | — | 8 | 40.524 | — | — | |
| PCA2026.03 | — | — | 30.5 | 1.301 | — | — | |
| PCNNFilter number=4000, Parameter number=5172k, FLOPs (M)=0.3822022.06 | — | — | 7.812 | — | — | — | |
| Random Reshuffle (RR)LR=Fixed, Optimizer=Adam, Epochs=1002026.03 | — | — | — | — | — | 0.0232 | |
| Random Reshuffle (RR)LR=Diminishing, Optimizer=Adam, Epochs=1002026.03 | — | — | — | — | — | 0.0223 | |
| ReLUFilter number=4000, Parameter number=5172k, FLOPs (M)=0.5582022.06 | — | — | 7.831 | — | — | — | |
| Robust CNNSetting=Adversarial Training, Training Attack=BIM2025.10 | — | — | — | — | 25.9 | — | |
| Robust CNN (BIM Train)Model=Robust CNN (BIM Train)2025.10 | — | — | 25.9 | — | — | — | |
| SELUFilter number=4000, Parameter number=5172k, FLOPs (M)=0.5582022.06 | — | — | 8.325 | — | — | — | |
| Shuffle Once (SO)LR=Fixed, Optimizer=Adam, Epochs=1002026.03 | — | — | — | — | — | 0.038 | |
| Shuffle Once (SO)LR=Diminishing, Optimizer=Adam, Epochs=1002026.03 | — | — | — | — | — | 0.0406 | |
| SigmoidFilter number=4000, Parameter number=5172k, FLOPs (M)=0.5582022.06 | — | — | 8.76 | — | — | — | |
| SIR2026.03 | — | — | 29.2 | 1.125 | — | — | |
| STEFilter number=4000, Parameter number=5172k, FLOPs (M)=0.3822022.06 | — | — | 7.797 | — | — | — | |
| SwishFilter number=4000, Parameter number=5172k, FLOPs (M)=0.5582022.06 | — | — | 8.285 | — | — | — | |
| t-SNE2026.03 | — | — | 9.2 | 220.947 | — | — | |
| TanhFilter number=4000, Parameter number=5172k, FLOPs (M)=0.5582022.06 | — | — | 8.165 | — | — | — | |
| UMAP Supervised2026.03 | — | — | 1.8 | 18.909 | — | — | |
| UMAP Unsupervised2026.03 | — | — | 14.5 | 13.922 | — | — |