AI 용어집
인공지능 완전 사전
Prototype Learning
Supervised learning where the model builds and uses representative examples (prototypes) to classify new instances based on their similarity.
Case-Based Reasoning
Problem-solving methodology that uses similar past cases as analogies to solve new problems, relying on memory of experiences.
Nearest Prototype Classifier
Classification algorithm that assigns a label to a new instance by comparing it to the nearest prototype in the feature space according to a defined distance metric.
Prototype Selection
Algorithmic process of identifying and retaining the most representative examples of a dataset while eliminating redundancies to optimize performance and interpretability.
Concept Activation Vectors
Directional vectors in the latent space of a neural network that capture the direction of maximum variation for a specific semantic concept, enabling the identification of conceptual prototypes.
K-Means Prototypes
Use of centroids resulting from the K-Means algorithm as representative prototypes of clusters, allowing predictions to be explained by proximity to class centroids.
Medoids
Real points of a cluster that minimize the sum of distances to all other points in the cluster, serving as more robust prototypes than centroids because they actually belong to the data.
Prototype-based Neural Networks
Neural network architectures that explicitly incorporate learnable prototype layers, enabling decisions based on similarity with representative examples.
Similarity Metrics
Mathematical functions quantifying the resemblance between two instances in the feature space, fundamental for identifying and comparing prototypes in explanatory systems.
Prototype Space
Dimensional subspace of the feature space where prototypes are projected and organized to facilitate visual interpretation and analysis of relationships between representative examples.
Critical Prototypes
Subset of prototypes having the greatest discriminatory impact on decision boundaries, identified by their maximum contribution to the separation between different classes.
Prototype Diversity
Measure of intra-class dissimilarity between selected prototypes, ensuring representative and comprehensive coverage of data variability in each category.
Prototype Relevance
Score quantifying the importance of a prototype for explaining a specific prediction, generally calculated as the inverse of the weighted distance between the instance and the prototype.
Interpretability by Examples
Explainability paradigm where predictions are interpreted through the presentation of similar examples and relevant counter-examples, making decisions understandable by analogy.
Instance-based Explanation
Explanatory approach that justifies a prediction by identifying and presenting the most similar training instances that influenced the model's decision.
Prototype Distance
Quantitative measure of the gap between an instance to be explained and its reference prototypes, used as the main criterion for generating similarity-based explanations.
Prototype Clustering
Hierarchical clustering technique of instances to automatically identify representative candidates of the different intra-class data structures.
Case-Based Explanation
Explanatory method adapted from case-based reasoning, where decisions are justified by presenting similar historical cases with their known and interpretable outcomes.