Research papers · 2022.10
Unifying Evaluation of Machine Learning Safety Monitors
Unifying Evaluation of Machine Learning Safety Monitors
Overview
The paper introduces Safety Gain, Residual Hazard, and Availability Cost to evaluate ML safety monitors. Examples including autonomous driving show how explicit reward, hazard, and system assumptions change conclusions about monitor effectiveness.
Author-supplied source information
Author-supplied information on arXiv.
9 pages, 5 figures, 3 tables, to appear in the proceedings of the 33rd IEEE International Symposium on Software Reliability Engineering (ISSRE 2022)
Version: arXiv v1 — manuscript revision