Image Classification on MNIST (test) (Accuracy, Weight Usage, and Depth Analysis)
98.4Accuracy (Full)BCNN-ISLaB-LRT
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| BCNN-ISLaB-LRTNumber of hidden layers=2, Neurons per layer=600, Initial weights=1’314’6402025.03 | 98.4 | 98.3 | — | — | — | |
| BNN-CONCRETENumber of hidden layers=2, Neurons per layer=600, Initial weights=1’314’6402025.03 | 98.4 | — | 90,139 | 3 | 3 | |
| BCNN-ISLaB-FLOWNumber of hidden layers=2, Neurons per layer=600, Initial weights=1’314’6402025.03 | 98.3 | 98.2 | — | — | — | |
| IS-ANN-L1Regularization strength (λ)=0.01, Number of hidden layers=2, Neurons per layer=600, Initial weights=1’314’6402025.03 | 98.3 | 98.4 | 12,276 | 1.91 | 3 | |
| BNN-HORSENumber of hidden layers=2, Neurons per layer=600, Initial weights=1’314’6402025.03 | 98.3 | 98 | 29,342 | 3 | 3 | |
| Laplace-SpaMNumber of hidden layers=2, Neurons per layer=600, Initial weights=1’314’6402025.03 | 98.1 | 98 | 22,608 | 3 | 3 | |
| ISLaB-LRTNumber of hidden layers=2, Neurons per layer=600, Initial weights=1’314’6402025.03 | 96.8 | 96.5 | 1,157.5 | 1.94 | 3 | |
| ISLaB-FLOWNumber of hidden layers=2, Neurons per layer=600, Initial weights=1’314’6402025.03 | 96.7 | 96.7 | 886.5 | 1.65 | 3 | |
| IS-ANN-L1Regularization strength (λ)=0.1, Number of hidden layers=2, Neurons per layer=600, Initial weights=1’314’6402025.03 | 92.7 | 92.4 | 2,196 | 1 | 1 | |
| IS-ANN-L1Regularization strength (λ)=0.2, Number of hidden layers=2, Neurons per layer=600, Initial weights=1’314’6402025.03 | 92.3 | 92.2 | 1,839.5 | 1 | 1 | |
| BLR-LRTNumber of hidden layers=2, Neurons per layer=600, Initial weights=1’314’6402025.03 | 92 | 91.9 | 593 | 1 | 1 | |
| BLR-FLOWNumber of hidden layers=2, Neurons per layer=600, Initial weights=1’314’6402025.03 | 91.9 | 91.9 | 606 | 1 | 1 | |
| IS-ANN-L1Regularization strength (λ)=1, Number of hidden layers=2, Neurons per layer=600, Initial weights=1’314’6402025.03 | 90.6 | 90.6 | 1,092.5 | 1 | 1 |