Authors: C. Tyler Diggans and A. R. AlMomani
This paper extends the Geometric Partition Entropy framework to estimate mutual information in settings where informative outliers are present. Traditional estimators either fail in the presence of outliers or aggressively remove them, discarding potentially critical signal. The generalized GPE approach preserves outlier information while maintaining robust estimation.
Explainable & Interpretable ML / AI (Pillar 2)