AI-woordenlijst
Het complete woordenboek van kunstmatige intelligentie
Weighted Density
Selection method combining model uncertainty measurement with a local density estimate to prioritize samples that are both uncertain and located in dense regions of the feature space.
Uncertainty Sampling
Active learning strategy that selects samples for which the model exhibits the lowest confidence in its predictions, generally measured by entropy or decision margin.
Query by Committee
Active learning approach using multiple models forming a committee, where samples causing the most disagreement among committee members are selected for annotation.
Diversity-Based Sampling
Selection strategy seeking to maximize the diversity of annotated samples to effectively cover the feature space and avoid information redundancy.
High-Density Points
Samples located in regions of the feature space with high data concentration, considered representative of the underlying data distribution.
Mutual Information Criterion
Informational utility metric measuring the expected reduction in uncertainty on model parameters after annotating a specific sample.
Confidence Margin
Difference between the predicted probabilities of the two most likely classes for a sample, used as an uncertainty indicator in active learning strategies.
Hybrid Selection
Approach combining multiple selection criteria (uncertainty, density, diversity) through weighting or multi-objective optimization to identify the most informative samples.
Outliers in Active Learning
Atypical or anomalous data points that density-based strategies seek to avoid, as their annotation provides little information about the general structure of the data.
Kernel Weighting
Technique using kernel functions to estimate local density and weight the importance of samples according to their similarity with their neighbors in the feature space.
Data Representativeness
Quality of a sample or subset to capture the essential characteristics of the overall data distribution, a key factor in effective sampling strategies.
Feature Spaces
Multidimensional domain where each dimension represents a feature of the data, used to analyze similarity and density relationships between samples.
Density-Uncertainty Criterion
Utility function combining a model uncertainty measure with a local density estimate to evaluate the information potential of each unlabeled sample.
Multi-Objective Optimization
Mathematical framework enabling the simultaneous handling of multiple conflicting objectives such as uncertainty, density, and diversity in active selection strategies.