Out-of-Distribution Detection on MNIST vs KMNIST (test)
1AUROCDE
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DEMemory budget=10p2024.09 | 1 | — | |
| LLA-DMemory budget=10p2024.09 | 0.98 | — | |
| LLA-DModel Architecture=MLP, Parameters (p)=15K, Memory Budget=3p2024.09 | 0.98 | — | |
| DIMSBackbone=LeNet, Hessian Approximation=low-rank, Parameter samples=100, friction coefficient=0.52026.05 | 0.979 | — | |
| Deep EnsembleModel Architecture=MLP, Parameters (p)=15K, Memory Budget=3p2024.09 | 0.96 | — | |
| RLABackbone=LeNet, Hessian Approximation=low-rank, Parameter samples=1002026.05 | 0.882 | — | |
| LINLABackbone=LeNet, Hessian Approximation=low-rank, Parameter samples=1002026.05 | 0.775 | — | |
| LABackbone=LeNet, Hessian Approximation=low-rank, Parameter samples=1002026.05 | 0.681 | — | |
| Sketched Lanczos UncertaintyMemory budget=10p2024.09 | 0.46 | — | |
| Sketched Lanczos UncertaintyModel Architecture=MLP, Parameters (p)=15K, Memory Budget=3p2024.09 | 0.42 | — | |
| SCODMemory budget=10p2024.09 | 0.36 | — | |
| LLAMemory budget=10p2024.09 | 0.34 | — | |
| LEMemory budget=10p2024.09 | 0.34 | — | |
| LE-HMemory budget=10p2024.09 | 0.34 | — | |
| SCODModel Architecture=MLP, Parameters (p)=15K, Memory Budget=3p2024.09 | 0.31 | — | |
| LLAModel Architecture=MLP, Parameters (p)=15K, Memory Budget=3p2024.09 | 0.3 | — | |
| LEModel Architecture=MLP, Parameters (p)=15K, Memory Budget=3p2024.09 | 0.3 | — | |
| LE-HModel Architecture=MLP, Parameters (p)=15K, Memory Budget=3p2024.09 | 0.3 | — | |
| SWAGModel Architecture=MLP, Parameters (p)=15K, Memory Budget=3p2024.09 | 0.19 | — | |
| SWAGMemory budget=10p2024.09 | 0.18 | — | |
| DAPPrBackbone=ConvNet (3 conv + 3 dense)2026.05 | — | 0.9881 | |
| DUQBackbone=ConvNet (3 conv + 3 dense)2026.05 | — | 0.9852 | |
| EDLBackbone=ConvNet (3 conv + 3 dense)2026.05 | — | 0.9631 | |
| F-EDLBackbone=ConvNet (3 conv + 3 dense)2026.05 | — | 0.9874 | |
| I-EDLBackbone=ConvNet (3 conv + 3 dense)2026.05 | — | 0.9833 | |
| KL-PNBackbone=ConvNet (3 conv + 3 dense)2026.05 | — | 0.9339 | |
| MC DropoutBackbone=ConvNet (3 conv + 3 dense)2026.05 | — | 0.94 | |
| PostNetBackbone=ConvNet (3 conv + 3 dense)2026.05 | — | 0.9459 | |
| R-EDLBackbone=ConvNet (3 conv + 3 dense)2026.05 | — | 0.9869 | |
| RKL-PNBackbone=ConvNet (3 conv + 3 dense)2026.05 | — | 0.5376 |