Uncertainty in the Variational Information Bottleneck
arXiv:1807.00906v1112 citations
Originality Synthesis-oriented
AI Analysis
This addresses uncertainty handling in machine learning models, but it is incremental as it applies an existing method to new tasks.
The paper demonstrates that the Variational Information Bottleneck (VIB) improves classification calibration and out-of-distribution detection in networks, providing natural metrics for uncertainty quantification without explicit design.
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.