Glosarium AI
Kamus lengkap Kecerdasan Buatan
Prototype Critique
Representative example that negatively influences a model's prediction, identified to understand cases where the model fails or produces counter-intuitive results.
k-Prototypes Method
Hybrid clustering algorithm combining k-means for numerical variables and k-modes for categorical variables, used to identify prototypes in mixed data.
Counterfactual Example
Minimally modified instance that changes the model's prediction, complementary to prototypes by explaining what would need to be changed to obtain a different result.
Prototype Similarity Map
Visualization showing how a new prediction relates to learned prototypes, facilitating the interpretation of model decisions by analogy.
Density-Based Prototype Selection
Method identifying prototypes in high-density regions of the data space, ensuring that representative examples are statistically significant.
MMD Method (Maximum Mean Discrepancy)
Statistical test used to measure the difference between distributions, applied to prototype selection to ensure they adequately cover the data space.
Explanation by Analogy
Explanation paradigm where a prediction is justified by its similarity to known prototypes, making complex decisions more intuitive for humans.
Maximum Margin Prototype
Prototype selected to maximize the decision distance from class boundaries, providing the most representative and unambiguous examples.
Prototype Neural Network (ProtoNN)
Neural network architecture that explicitly learns a set of prototypes for classification, offering intrinsic interpretability by design.
Prototype Coreset
Minimal subset of prototypes preserving the geometric properties of the complete dataset, optimizing explanation efficiency without loss of relevance.
Medoid Method
Variant of k-means where cluster centers are actual data points (medoids), ensuring that prototypes are concrete and interpretable examples.
Feature Weighting by Prototype
Technique assigning weights to features specific to each prototype, clarifying which aspects are most important for each type of representative example.
Prototype Cross-Validation
Process of evaluating the robustness of prototype-based explanations by testing their stability on different subsets of training data.
Hybrid Prototype
Representative example combining characteristics from multiple real instances, created to synthesize the most relevant traits of a class or concept.
Elbow Method for Prototypes
Heuristic technique determining the optimal number of prototypes by identifying the point where adding new examples no longer significantly improves coverage.
Local Explanation by Prototype
Approach explaining an individual prediction by associating it with the most similar prototype, providing contextual and case-specific justification.