Out-of-distribution detection on MNIST Gaussian
100AUROCAENIB
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| AENIBAugmentation=random translation2023.03 | 100 | — | — | — | — | — | — | |
| DisenIBAugmentation=random translation2023.03 | 99.7 | — | — | — | — | — | — | |
| VIBAugmentation=random translation2023.03 | 98.8 | — | — | — | — | — | — | |
| Cross-entropyAugmentation=random translation2023.03 | 98.7 | — | — | — | — | — | — | |
| NLIBAugmentation=random translation2023.03 | 97.4 | — | — | — | — | — | — | |
| sq-VIBAugmentation=random translation2023.03 | 95.5 | — | — | — | — | — | — | |
| sq-NLIBAugmentation=random translation2023.03 | 90.8 | — | — | — | — | — | — | |
| Softmax BaselineArchitecture=3-layer FC (256-wide), Nonlinearity=GELU, Optimizer=Adam, Training Epochs=302016.10 | 90 | 50 | 90 | 50 | 91 | 50 | 91 |