AI-ordlista
Den kompletta ordlistan över AI
Margin of Uncertainty
Measure calculated as the difference between the probabilities of the two most likely classes, used to identify samples where the model hesitates between predictions.
Prediction Entropy
Uncertainty metric based on Shannon entropy applied to the probability distribution of the model's predictions, favoring samples with uniform distributions.
Prediction Variance
Uncertainty measure quantifying the variability of the model's predictions, often calculated via techniques such as Monte Carlo Dropout or model ensembles.
Least Confidence
Simple strategy selecting samples where the model has the lowest confidence in its most probable prediction, identified by the lowest maximum probability.
Entropy Sampling
Method selecting samples that maximize the entropy of the predicted probability distribution, indicating maximum model uncertainty.
Density-weighted Uncertainty
Approach combining model uncertainty with data density in the feature space to avoid selecting outlier samples.
BALD
Bayesian Active Learning by Disagreement, method using mutual information to measure the expected disagreement between model predictions and posterior parameters.
Monte Carlo Dropout
Bayesian approximation technique using dropout during inference to estimate model uncertainty through multiple stochastic predictions.
Annotation Budget
Constraint defining the maximum number of samples that can be labeled, guiding selection strategies to maximize learning efficiency.