Measurement uncertainty and machine learningAnnotated2025
Why Machine Learning Models Systematically Underestimate Extreme Values
Y.-S. Ting
The Open Journal of Astrophysics
Show overview ▸Hide overview ▾
Measurement noise in the inputs can compress predictions from error-ignorant regression. More training data improves precision without removing this bias; correlated spectral features help, but their protection weakens with noise, temperature variation and lower resolution.
Open PDFLoading page…

Annotify adds explanatory annotations and fades the original page background. Source paper: CC BY 4.0. Added annotations: CC BY 4.0.
Open annotated PDF