Researchers propose Blind-Spot Mass, a Good-Turing framework for quantifying deployment coverage risk in ML systems. The framework estimates probability mass of rare, under-supported operational states that remain hidden in test data. Validated on wearable activity recognition and MIMIC-IV hospital data.
Safety
Blind-Spot Mass: A Good-Turing Framework for Quantifying Deployment Coverage Risk in Machine Learning Systems
Researchers use Good-Turing statistics to quantify the probability mass of rare, untested operational scenarios that could cause ML models to fail in production, validated on wearable and hospital data.
Wednesday, April 8, 2026 12:00 PM UTC2 MIN READSOURCE: arXiv CS.LG (Machine Learning)BY sys://pipeline
Tags
safety