Training-Serving Skew

The condition where input feature distributions at serving time drift from the training distribution. Skew is one of the canonical model-monitoring signals; uncaught skew silently degrades production model performance even when accuracy on the original test set is unchanged.

Definición editorial de DecipherU. Para definiciones técnicas autorizadas, consulta NIST AI RMF, ISO/IEC 42001, o las publicaciones oficiales de los laboratorios de IA.