List your company
ResearchFormerly WizWand

Classification on MNIST 10,000 samples (test)

4.2Test ErrorBest baseline (SVM/MLP)
3.7326.89110.0513.209Apr 5, 2026
Updated 6mo ago

Evaluation results

MethodLinks
Best baseline (SVM/MLP)PCA dimension=64, Training size (N)=10,0002026.044.2—
Best baseline (MLP/SVM)PCA dimension=32, Training size (N)=10,0002026.044.6—
Best baseline (MLP/SVM)PCA dimension=32, Training size (N)=5,0002026.045.7—
Best baseline (SVM/MLP)PCA dimension=64, Training size (N)=5,0002026.045.8—
Best baseline (SVM/MLP)PCA dimension=16, Training size (N)=10,0002026.046—
Best baseline (SVM/MLP)PCA dimension=16, Training size (N)=5,0002026.046.9—
ArrowFlow bestPCA dimension=64, Training size (N)=10,000, Architecture=[1024,128] 3-layer, Ensemble=7-view2026.049.14.9
ArrowFlow bestPCA dimension=32, Training size (N)=10,000, Architecture=[1024,128] 3-layer, Ensemble=7-view2026.049.34.7
ArrowFlow bestPCA dimension=16, Training size (N)=10,000, Architecture=[1024,128] 3-layer, Ensemble=7-view2026.049.83.8
Best baseline (MLP/SVM)PCA dimension=32, Training size (N)=1,0002026.0410.5—
Best baseline (SVM/MLP)PCA dimension=64, Training size (N)=1,0002026.0410.8—
Best baseline (SVM/MLP)PCA dimension=16, Training size (N)=1,0002026.0411.2—
ArrowFlow bestPCA dimension=64, Training size (N)=5,000, Architecture=[1024,128] 3-layer, Ensemble=7-view2026.0411.25.4
ArrowFlow bestPCA dimension=32, Training size (N)=5,000, Architecture=[1024,128] 3-layer, Ensemble=7-view2026.0411.45.7
ArrowFlow bestPCA dimension=16, Training size (N)=5,000, Architecture=[1024,128] 3-layer, Ensemble=7-view2026.0412.25.3
ArrowFlow bestPCA dimension=64, Training size (N)=1,000, Architecture=[1024,128] 3-layer, Ensemble=7-view2026.0415.24.4
ArrowFlow bestPCA dimension=32, Training size (N)=1,000, Architecture=[1024,128] 3-layer, Ensemble=7-view2026.0415.65.1
ArrowFlow bestPCA dimension=16, Training size (N)=1,000, Architecture=[1024,128] 3-layer, Ensemble=7-view2026.0415.94.7