Unifying Runtime Monitoring Approaches for Safety-Critical Machine Learning: Application to Vision-Based Landing
A new paper proposes a unified framework to categorize runtime monitoring approaches for safety-critical machine learning applications. The framework divides monitoring into three types: Operational Design Domain (ODD) monitoring, Out-of-Distribution (OOD) monitoring, and Out-of-Model-Scope (OMS) monitoring. This categorization aims to streamline the design, evaluation, and comparison of different monitoring methods, with a practical demonstration on runway detection for aeronautical safety. AI
IMPACT Provides a structured approach to enhance the safety and reliability of ML systems in critical applications.