Research papers · Posted 2021.10.04 · Publication / acceptance: author-reported
Benchmarking Safety Monitors for Image Classifiers with Machine Learning
Benchmarking Safety Monitors for Image Classifiers with Machine Learning
Overview
The framework benchmarks image-classifier safety monitors from data generation through evaluation. Tests across novelty, noise, anomalies, distribution shifts, and adversarial examples reveal limitations in monitor accuracy under the chosen benchmark conditions.
Publication and version information
Publication, acceptance or venue information is author-reported on arXiv. The exact source wording is preserved below.
Author-supplied source information
Author-supplied information on arXiv.
26th IEEE Pacific Rim International Symposium on Dependable Computing (PRDC 2021), IEEE, Dec 2021, Perth, Australia
Version: arXiv v1 — manuscript revision