AI-ordlista
Den kompletta ordlistan över AI
Weighted Sample
A training sample associated with a weight indicating its relative importance in the algorithm, with these weights being adjusted iteratively based on classification errors.
Error Rate
The proportion of samples misclassified by a classifier, used in AdaBoost to calculate the classifier's weight in the final decision and to adjust the sample weights.
Alpha Coefficient
The weight assigned to each weak classifier in the final AdaBoost model, calculated from its error rate using the formula α = 0.5 * ln((1-error)/error).
Exponential Loss
The loss function used by AdaBoost that exponentially penalizes classification errors, thereby encouraging a rapid correction of misclassified samples.
Sample Weight Update
The iterative process in AdaBoost where the weights of samples are increased for misclassified ones and decreased for correctly classified ones, forcing the next learner to focus on difficult samples.
Voting Weight
The multiplicative coefficient applied to the prediction of each weak classifier during the final weighted vote in AdaBoost, based on its individual performance.
Adaptive Boosting
The fundamental principle of AdaBoost where the algorithm dynamically adapts its strategy by focusing on the most difficult samples at each iteration.
Iterative Training
A sequential training process where each new model is trained on a modified version of the original dataset, taking into account the performance of previous models.
Classification Margin
Measure of confidence of a classification calculated as the difference between the cumulative weights of classifiers voting for the correct class and those voting for the incorrect class.
Overfitting in Boosting
Phenomenon where an AdaBoost model overfits the training data, particularly when the number of iterations is too high or the weak learners are too complex.