AI-ordlista
Den kompletta ordlistan över AI
Pasting Ensemble
Ensemble method that builds multiple models on random subsets of the training data, without replacement, to reduce variance and improve generalization.
Sampling without Replacement
Observation selection technique where each chosen element is removed from the population, ensuring unique subsets as in pasting.
Sampling with Replacement
Method where observations can be selected multiple times in the same sample, a fundamental characteristic of bagging.
Training Subset
Portion of the training data used to build an individual model in an ensemble method, with or without replacement depending on the technique.
Prediction Aggregation
Process of combining individual predictions from ensemble models, typically by majority vote (classification) or averaging (regression).
Model Diversity
Principle that ensemble models must be different for aggregation to be effective, achieved through varied data subsets.
Random Subspace Sampling
Extension of bagging where models are trained on random subsets of features in addition to observation subsets.
Pasting Small Samples
Pasting variant using reduced-size subsets to speed up training while maintaining model diversity.
Model Variance
Model sensitivity to variations in training data, which ensemble methods like bagging specifically aim to reduce.
Prediction Stability
A model's ability to produce consistent predictions in the face of slight variations in training data, improved by ensemble methods.
Parallel Ensemble Training
Advantage of bagging and pasting allowing simultaneous training of base models on different data subsets.
Sample Complexity
Number of samples needed to achieve a certain performance, potentially reduced by effective ensemble methods.