AI-ordlista
Den kompletta ordlistan över AI
SHAP Value
Explainability method based on game theory that attributes to each feature a marginal contribution to the final prediction while respecting the axioms of efficiency and symmetry.
Permutation Feature Importance
Technique measuring the decrease in model performance when the values of a variable are randomly permuted, thus quantifying its predictive contribution.
Partial Dependence Analysis
Visualization showing the marginal effect of one or two variables on the average prediction while marginalizing the effect of other features.
Integrated Gradients
Attribution method calculating the integral of gradients along a path from a reference to the input to determine the importance of each feature.
Relevance Map
Visualization indicating the regions of input data that most influence the model's prediction, particularly used in computer vision.
Surrogate Decision Tree
Simple interpretable model trained to approximate the behavior of a complex model, allowing for a global understanding of decision rules.
Robustness Analysis
Evaluation of the stability of model predictions in the face of input data perturbations to identify vulnerability points.
Local Sensitivity Analysis
Detailed examination of the impact of input variations on a specific prediction to interpret individual model decisions.
Sobol indices
Metrics decomposing the output variance into contributions from input variables and their interactions to quantify the influence of each factor.
Morris test
Efficient screening method identifying input variables with significant linear, nonlinear, or interaction effects on predictions.
Uncertainty analysis
Quantification of uncertainties in model predictions due to input data variability and limitations of the model itself.
Input perturbation
Technique introducing controlled variations in input data to assess the sensitivity and stability of model predictions.
Functional ANOVA decomposition
Analytical approach decomposing the model function into main effects and interaction terms to precisely quantify contributions.
Conditional dependence analysis
Study of relationships between variables conditioned by other characteristics to reveal complex and nonlinear dependencies.
Instance influence
Measure quantifying the impact of a specific data point on model parameters and predictions during its training.
FAST method
Global sensitivity analysis approach using Fourier transform to efficiently estimate sensitivity indices of all orders.