VNDUQE uses VIB on MNIST to achieve 95.3% AUROC for OOD detection by combining KL divergence for far-OOD and entropy for near-OOD, outperforming MSP baseline by 10 points while reducing calibration error.
PyTorch: An Imperative Style, High-Performance Deep Learning Library
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
VNDUQE: Information-Theoretic Novelty Detection using Deep Variational Information Bottleneck
VNDUQE uses VIB on MNIST to achieve 95.3% AUROC for OOD detection by combining KL divergence for far-OOD and entropy for near-OOD, outperforming MSP baseline by 10 points while reducing calibration error.