AI-woordenlijst
Het complete woordenboek van kunstmatige intelligentie
Softmax Calibration
Process of adjusting probabilities generated by the softmax function to accurately reflect the true occurrence probabilities of predictions.
Post-hoc Calibration
Calibration approach applied after complete model training, without modifying its weights or architecture, using only the output logits.
Sharpness vs Calibration Trade-off
Fundamental dilemma where improving calibration may reduce the sharpness (confidence) of predictions, and vice versa, requiring optimal balance.
Bayesian Calibration
Probabilistic calibration approach incorporating parameter uncertainty and using Bayesian methods to estimate calibrated posterior distributions.
Ensemble Calibration
Technique improving calibration by aggregating predictions from multiple models, naturally reducing overconfidence and improving uncertainty estimation.