Uncertainty in the Variational Information Bottleneck
About
We present a simple case study, demonstrating that Variational Information Bottleneck (VIB) can improve a network's classification calibration as well as its ability to detect out-of-distribution data. Without explicitly being designed to do so, VIB gives two natural metrics for handling and quantifying uncertainty.
Alexander A. Alemi, Ian Fischer, Joshua V. Dillon• 2018
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Out-of-Distribution Detection | CIFAR-10 (ID) vs SVHN (OOD) (test) | AUROC52.8 | 79 | |
| Out-of-Distribution Detection | FashionMNIST (ID) vs MNIST (OoD) | AUROC0.941 | 61 | |
| Out-of-Distribution Detection | CIFAR-10 (ID) vs Celeb-A (OOD) | AUROC73.5 | 55 | |
| OOD Detection | FashionMNIST (In-Distribution) vs Omniglot (Out-of-Distribution) original (test) | AUROC0.943 | 4 |
Showing 4 of 4 rows