AI-ordlista
Den kompletta ordlistan över AI
Model Selection
Systematic process of identifying and selecting the best performing models within a training set. This selection is based on criteria of performance, diversity, and complementarity between models.
Diversity-based Pruning
Pruning method that favors the selection of models with diverse and complementary predictions. The goal is to maximize the diversity of the final ensemble to improve the robustness of predictions.
Greedy Pruning
Iterative pruning algorithm that selects or eliminates models one by one according to a local optimization criterion. This method seeks a suboptimal but computationally efficient solution.
Genetic Pruning
Pruning approach inspired by genetic algorithms that uses selection, crossover, and mutation operators to optimize the ensemble composition. This method explores the solution space more exhaustively.
Clustering-based Pruning
Pruning technique that groups similar models into clusters and then selects representatives from each cluster. This method ensures good coverage of the prediction space while reducing redundancy.
Ensemble Selection
Process of optimally choosing a subset of models from a larger ensemble to maximize prediction performance. This selection can be formulated as a combinatorial optimization problem.
Model Ranking
Establishing a priority order of models in the ensemble based on performance and contribution metrics. This ranking guides the pruning process by identifying the most relevant models.
Ensemble Reduction
Controlled reduction of the size of a training ensemble while preserving its generalization capabilities. This reduction aims to find an optimal trade-off between performance and computational complexity.
Pruning Strategy
Set of rules and algorithms defining how models are evaluated and eliminated during the pruning process. The strategy determines the selection criteria and the order of model elimination.
Subset Selection
Optimization problem consisting of identifying the best subset of models that maximizes a given objective function. This selection can be approached through exact or heuristic methods.
Ensemble Optimization
Process of improving the composition and structure of an ensemble to achieve optimal performance. Optimization includes pruning, weighting, and selection of ensemble members.
Pruning Threshold
Critical value used as a decision criterion to eliminate models during the pruning process. This threshold can be based on performance, diversity, or marginal contribution metrics.
Ensemble Compression
Reduction of an ensemble's size while preserving its predictive capabilities through distillation or pruning techniques. This compression facilitates deployment and inference in production.
Forward Selection
Constructive pruning method that progressively adds the most beneficial models to the ensemble. This approach ensures that each added model contributes positively to overall performance.
Backward Elimination
Pruning strategy that starts with the complete ensemble and iteratively eliminates the least useful models. This method allows evaluating the impact of each model on the ensemble's performance.
Pruning Criteria
Set of metrics and rules used to evaluate the usefulness of each model in the ensemble. These criteria typically combine performance, diversity, and complexity measures.
Ensemble Simplification
Processus de réduction de la complexité structurelle d'un ensemble tout en maintenant ses capacités prédictives. Cette simplification peut inclure l'élagage, la fusion de modèles similaires et la réduction des paramètres.