While deep learning models have seen widespread success in controlled environments, there are still barriers to their adoption in open-world settings. One critical task for safe deployment is the detection of anomalous or out-of-distribution samples that may require human intervention. In this work, we present a novel loss function and recipe for training networks with improved density-based out-of-distribution sensitivity. We demonstrate the effectiveness of our method on CIFAR-10, notably reducing the false-positive rate of the relative Mahalanobis distance method on far-OOD tasks by over 50%.
@article{arxiv.2311.00808,
title = {Mahalanobis-Aware Training for Out-of-Distribution Detection},
author = {Connor Mclaughlin and Jason Matterer and Michael Yee},
journal= {arXiv preprint arXiv:2311.00808},
year = {2023}
}
Comments
2 pages, 2 figures. Presented at AAAI Fall Symposium Series `23