Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Conformal Prediction
Statistical framework that produces prediction sets with valid coverage guarantees under minimal assumptions, transforming point predictions into confidence intervals.
Split Conformal Prediction
Variant of conformal prediction that divides data into training and calibration sets, ensuring the necessary independence for validity of prediction sets.
Conformalized Quantile Regression
Method combining quantile regression and conformal prediction to construct prediction bands with guaranteed coverage in regression.
Conformalized Risk Control
Extension of conformal prediction to control the FWER (Family-Wise Error Rate) risk in multiple variable selection problems.
Adaptive Conformal Prediction
Approach that dynamically adjusts conformity levels based on data characteristics to optimize the size of prediction sets.
Conformal Prediction Sets
Possible label sets generated by conformal prediction that contain the true label with a guaranteed probability of 1-α.
Coverage Guarantee
Mathematical property ensuring that prediction sets contain the true value with probability at least equal to the nominal confidence level.
Conformity Score
Function measuring how well a new observation conforms to the training data, fundamental for determining prediction sets.
P-value Conformal
p-values calculated from conformity scores to evaluate the inclusion of each possible label in the prediction set.
Mondrian Conformal Prediction
Extension of conformal prediction using partitions of the feature space to obtain conditional coverage guarantees.
Venn-Abers Predictors
Conformal prediction method for binary classification producing calibrated probabilities with frequentist validity guarantees.
Conformalized Residual Networks
Neural network architecture integrating conformal prediction mechanisms to produce predictions with quantified uncertainty.
Inductive Conformal Prediction
Variant of conformal prediction using a model pre-trained on a training set distinct from the calibration set.
Transductive Conformal Prediction
Original approach of conformal prediction retraining the model for each new observation by including it in the calibration process.
Conformal Calibration
Process of adjusting predictions to ensure that predicted probabilities correspond to observed empirical frequencies.
Jackknife Conformal Prediction
Conformal prediction method using the jackknife technique to estimate conformity scores without explicit data partitioning.
Conformalized Feature Selection
Application of conformal prediction to variable selection to control the false discovery rate with statistical guarantees.
Localized Conformal Prediction
Adaptive approach that adjusts prediction sets locally according to point density and prediction difficulty.