Image Classification on MNIST (test) (Accuracy and Calibration Metrics)
97.7AccuracyVaLLA 100
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| VaLLA 100Inducing points count=1002026.06 | 97.7 | 0.076 | 1 | 0.036 | |
| VaLLA 200Inducing points count=2002026.06 | 97.7 | 0.075 | 1 | 0.035 | |
| MAP2026.06 | 97.6 | 0.076 | 0.8 | 0.036 | |
| LLA*Layer restriction=last-layer2026.06 | 97.6 | 0.081 | 1.5 | 0.037 | |
| LLA* KFACLayer restriction=last-layer, Covariance approximation=Kronecker2026.06 | 97.6 | 0.081 | 1.5 | 0.037 | |
| ELLARandom points=20002026.06 | 97.6 | 0.076 | 0.8 | 0.036 | |
| Sampled LLA2026.06 | 97.6 | 0.087 | 2.6 | 0.04 | |
| LLA KFACCovariance approximation=Kronecker2026.06 | 97.5 | 0.094 | 2.9 | 0.041 | |
| LLA DiagCovariance approximation=Diagonal2026.06 | 97.4 | 0.143 | 7.2 | 0.053 |