Image Classification on MNIST (test) (Accuracy, Calibration, and Efficiency)
0.03NLLDAK
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DAK2026.05 | 0.03 | 99.06 | 0.5 | 3 | 13.73 | 1.42 | |
| DKL-SIKA2026.05 | 0.03 | 99.06 | 0.53 | 3 | 10.09 | 1 | |
| ANN-L1Network Density=Full, lambda=0.012025.03 | 0.05 | — | 0.8 | — | — | — | |
| ANN-L1Network Density=Sparse, lambda=0.012025.03 | 0.055 | — | 0.8 | — | — | — | |
| Laplace-SpaMNetwork Density=Full2025.03 | 0.06 | — | 0.4 | — | — | — | |
| BCNN-ISLaB-LRTNetwork Density=Sparse2025.03 | 0.066 | — | 0.3 | — | — | — | |
| BNN-HORSENetwork Density=Full2025.03 | 0.067 | — | 0.8 | — | — | — | |
| Laplace-SpaMNetwork Density=Sparse2025.03 | 0.068 | — | 1.3 | — | — | — | |
| DGP 3Model Type=Deep Gaussian Process, Number of layers (L)=3, Source=Salimbeni and Deisenroth (2017)2026.06 | 0.072 | 97.86 | — | — | — | — | |
| BCNN-ISLaB-LRTNetwork Density=Full2025.03 | 0.072 | — | 0.6 | — | — | — | |
| BCNN-ISLaB-FLOWNetwork Density=Sparse2025.03 | 0.073 | — | 0.3 | — | — | — | |
| DVIP 2Model Type=Deep Variational Implicit Process, Number of layers (L)=22026.06 | 0.074 | 98.39 | — | — | — | — | |
| BNN-HORSENetwork Density=Sparse2025.03 | 0.078 | — | 1 | — | — | — | |
| DVIP 3Model Type=Deep Variational Implicit Process, Number of layers (L)=32026.06 | 0.08 | 98.36 | — | — | — | — | |
| DGP 2Model Type=Deep Gaussian Process, Number of layers (L)=2, Source=Salimbeni and Deisenroth (2017)2026.06 | 0.082 | 97.75 | — | — | — | — | |
| BCNN-ISLaB-FLOWNetwork Density=Full2025.03 | 0.084 | — | 0.9 | — | — | — | |
| BNN-CONCRETENetwork Density=Full2025.03 | 0.086 | — | 1 | — | — | — | |
| SVDKL2026.05 | 0.09 | 98.17 | 1.87 | 16 | 16.46 | 1.81 | |
| ISLaB-FLOWNetwork Density=Sparse2025.03 | 0.129 | — | 0.5 | — | — | — | |
| ISLaB-FLOWNetwork Density=Full2025.03 | 0.135 | — | 0.5 | — | — | — | |
| VIPModel Type=Single-layer, Number of layers (L)=12026.06 | 0.144 | 97.99 | — | — | — | — | |
| SGPModel Type=Single-layer, Number of layers (L)=12026.06 | 0.146 | 96.25 | — | — | — | — | |
| ISLaB-LRTNetwork Density=Sparse2025.03 | 0.152 | — | 1 | — | — | — | |
| ISLaB-LRTNetwork Density=Full2025.03 | 0.161 | — | 1 | — | — | — | |
| ANN-L1Network Density=Full, lambda=0.12025.03 | 0.265 | — | 2.6 | — | — | — | |
| ANN-L1Network Density=Full, lambda=0.22025.03 | 0.276 | — | 3.5 | — | — | — | |
| ANN-L1Network Density=Sparse, lambda=0.12025.03 | 0.277 | — | 2.8 | — | — | — | |
| ANN-L1Network Density=Sparse, lambda=0.22025.03 | 0.286 | — | 3.6 | — | — | — | |
| BLR-LRTNetwork Density=Sparse2025.03 | 0.305 | — | 1.8 | — | — | — | |
| BLR-LRTNetwork Density=Full2025.03 | 0.311 | — | 1.9 | — | — | — | |
| BLR-FLOWNetwork Density=Sparse2025.03 | 0.316 | — | 1.9 | — | — | — | |
| BLR-FLOWNetwork Density=Full2025.03 | 0.327 | — | 2.1 | — | — | — | |
| ANN-L1Network Density=Full, lambda=12025.03 | 0.337 | — | 7.5 | — | — | — | |
| ANN-L1Network Density=Sparse, lambda=12025.03 | 0.357 | — | 7.7 | — | — | — |