Regression on Taxi (test)
3.112NLLGAPA
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| GAPA2026.02 | 3.112 | 4.035 | 10.4 | |
| VaLLA200config_parameter=2002026.02 | 3.232 | 3.979 | 14.2 | |
| VaLLA100config_parameter=1002026.02 | 3.235 | 3.999 | 14.9 | |
| VaLLA 200Number of inducing points=2002026.06 | 3.28 | 3.993 | 18.8 | |
| VaLLA 100Number of inducing points=1002026.06 | 3.287 | 3.968 | 18.8 | |
| Ensemble2026.02 | 3.631 | 3.384 | 96.1 | |
| LLA KFACLaplace approximation variant=Kronecker-factorized2026.06 | 3.705 | 3.977 | 27 | |
| LLA KFACHessian Approximation=KFAC2026.02 | 3.706 | 3.986 | 25.6 | |
| LLA DiagLaplace approximation variant=diagonal2026.06 | 3.714 | 3.979 | 27 | |
| LLA*Laplace approximation variant=last-layer2026.06 | 3.718 | 3.975 | 27 | |
| LLA* KFACLaplace approximation variant=last-layer Kronecker-factorized2026.06 | 3.718 | 3.976 | 27 | |
| LLA DiagHessian Approximation=Diagonal2026.02 | 3.722 | 3.99 | 25.7 | |
| LLA*Hessian Approximation=Diagonal, Variant=Last-layer2026.02 | 3.726 | 3.985 | 25.6 | |
| LLA*KFACHessian Approximation=KFAC, Variant=Last-layer2026.02 | 3.726 | 3.985 | 25.6 | |
| ELLALaplace approximation variant=accelerated2026.06 | 3.753 | 3.754 | 22.7 | |
| MAPbaseline=maximum-a-posteriori2026.06 | 3.763 | 3.753 | 22.7 | |
| MAP2026.02 | 3.775 | 3.755 | 21.1 | |
| Dropout2026.02 | 3.849 | 4.592 | 95.1 | |
| ELLA2026.02 | 3.885 | 3.68 | 21.9 |