YZ Sözlüğü
Yapay Zekanın tam sözlüğü
Counterfactual Actionability
The ability of a counterfactual explanation to suggest only modifications to features over which the end user has direct control, making the explanation directly useful for decision-making.
Counterfactual Latent Space
The compressed and nonlinear representation of data in which the search for counterfactual examples is performed, allowing for more efficient navigation and generation of more coherent explanations for complex models.
Counterfactual Adversarial Generation
An approach using generative adversarial networks (GANs) to produce counterfactual explanations, where a generator creates examples and a discriminator evaluates their validity and proximity to the original instance.
Counterfactual Causality Constraints
The incorporation of known cause-effect relationships into the generation process, ensuring that the proposed changes respect logical and temporal dependencies between features to avoid inconsistent explanations.
Multi-Output Counterfactual (MOC)
An extension of classical counterfactual explanations that aims to find a minimal input instance that can lead to several different alternative outputs, offering a richer view of change possibilities.
Group Counterfactual
An explanation that identifies the minimal modifications to apply to a set of instances sharing common characteristics to collectively change their prediction, useful for understanding decisions at the population level.
Multi-objective Counterfactual Optimization
The formulation of the search for counterfactual explanations as an optimization problem aiming to simultaneously balance multiple conflicting objectives, such as minimality, validity, plausibility, and diversity.
Instance-based Counterfactual
A method that generates explanations by identifying and slightly modifying real instances from the dataset that already have the desired prediction, thus ensuring high plausibility and grounding in data reality.
Counterfactual Feature Weighting
A technique that assigns different weights to features when calculating counterfactual distance, in order to prioritize modifying variables deemed easier or less costly to change in practice.
Hybrid Counterfactual
An approach combining instance-based and model-based methods to generate explanations, using proximity to real data for plausibility and surrogate models to explore more complex changes.