Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Partial Dependence Plot (PDP)
Plot illustrating the marginal effect of one or two features on the model's prediction, averaging predictions over all other features to reveal functional relationships.
Accumulated Local Effects (ALE)
Alternative to PDP that calculates the effect of a feature on the prediction by conditioning on the values of other features, thus avoiding biases due to correlations between variables.
Anchors Rules
High-fidelity explanation system that identifies a set of conditions (the anchor) on input features that sufficiently guarantee the model's prediction, making the explanation easily verifiable by an operator.
Saliency Map
Visualization technique, mainly for neural networks, that highlights input features (pixels, words, sensors) with the greatest influence on the final prediction via gradient calculation.
Force Plot
SHAP-specific visualization that shows how each feature pushes the prediction from the base value (average of predictions) towards the final predicted value, illustrating positive and negative contributions.
Formal Logical Explanation
XAI approach that extracts logical rules (if-then) from a black-box model to provide symbolic and understandable explanations, often used for compliance and security audits.