AI-ordlista
Den kompletta ordlistan över AI
Counterexample
Instance generated that differs minimally from the original input while producing a different model prediction, allowing to explain the conditions for decision change.
Hamming Distance
Metric measuring the minimum number of features to modify to transform an instance into a counterexample, essential for evaluating the parsimony of explanations.
Counterexample Validity
Fundamental criterion ensuring that the generated counterexample actually produces the desired and different prediction from that of the original instance.
Spatial Proximity
Similarity measure between the original instance and its counterexample in the feature space, crucial for ensuring the relevance and credibility of the explanation.
Actionability
Ability of a counterexample to suggest feasible and relevant modifications for the user, transforming the explanation into practical recommendations.
Counterexample Diversity
Generation of sets of counterexamples covering different alternative pathways to modify the prediction, offering a comprehensive view of intervention possibilities.
Model Robustness
Analysis of prediction stability in the face of counterexamples, revealing the model's weaknesses and its sensitivity areas to input variations.
Human-Machine Interface
Interactive system allowing users to generate and explore counterexamples, facilitating intuitive understanding of the model's decision mechanisms.
Causal Interpretation
Extension of counterexamples to identify cause-effect relationships between features and predictions, beyond mere statistical correlations.
Counterexample Set
Organized collection of counterexamples exploring different modification dimensions, providing a multidimensional understanding of decision boundaries.
Penalty Regularization
Mechanism controlling the magnitude of modifications allowed during counterexample generation, avoiding unrealistic or excessive changes.
Membership Constraints
Rules ensuring that counterexamples remain within valid and plausible domains, respecting the physical or logical boundaries of the problem.
Exploration Method
Systematic algorithm for efficiently discovering relevant counterexamples in the high-dimensional feature space of the model.
Cost Function
Mathematical expression combining multiple criteria to evaluate the quality of a counterexample, guiding optimization toward the most relevant solutions.