Conformal prediction
Conformal prediction: A framework constructing prediction sets or intervals with coverage guarantees under exchangeability assumptions. It uses nonconformity scores on a separate calibration set. Shift and dependence can invalidate marginal coverage or require specific variants. This perspective connects the definition with practical behaviour in artificial intelligence systems.
In italiano: Conformal PredictionLegge II · Probabilità, softmax e modelli generativiIt uses nonconformity scores on a separate calibration set. Shift and dependence can invalidate marginal coverage or require specific variants.