AI 용어집
인공지능 완전 사전
Gradient Boosting
Ensemble learning technique that builds predictive models sequentially, where each new model corrects the errors of the previous ones by optimizing a loss function via gradient descent.
Additive Learning
Fundamental principle of Gradient Boosting where the final model is the weighted sum of predictions from multiple weak learners, each added to improve the overall performance.
Learning Rate
Hyperparameter that controls the influence of each weak learner on the final model, acting as a weighting factor to prevent overfitting.
Residuals
Prediction errors of the current model, calculated as the difference between observed values and predictions, on which the next weak learner is trained in Gradient Boosting.
Number of Estimators
Hyperparameter defining the number of weak learners (iterations) to build in the Gradient Boosting model, directly influencing complexity and performance.
XGBoost (Extreme Gradient Boosting)
Optimized and parallelized implementation of Gradient Boosting that incorporates regularization, handling of missing values, and tree pruning techniques for superior efficiency.
LightGBM
Gradient Boosting framework that uses a leaf-wise tree growth technique instead of level-wise, offering increased training speed and reduced memory consumption.
CatBoost
Gradient Boosting algorithm specialized in the efficient handling of categorical features, using ordered encoding techniques and asymmetrical boosting schemes.
Stochastic Gradient Boosting
Variant of Gradient Boosting where each weak learner is trained on a random subset of training data, reducing correlation between trees and improving generalization.
Feature Subsampling
Regularization technique in Gradient Boosting that involves considering only a random subset of predictive variables for each tree node split, limiting overfitting.
Maximum Tree Depth
Hyperparameter controlling the complexity of each weak learner by limiting the number of decision splits, balancing bias and variance in Gradient Boosting models.
Pseudo-Residuals
Generalization of residuals in Gradient Boosting, representing the negative gradient of the loss function with respect to current predictions, enabling optimization for various loss functions.
Regression Boosting
Application of Gradient Boosting to regression problems where the goal is to predict continuous values, typically using a squared or absolute loss function.
Classification Boosting
Application of Gradient Boosting to classification problems, using specific loss functions like log-loss (cross-entropy) to guide optimization of class probabilities.
L1/L2 Regularization
Penalization techniques added to the loss function in Gradient Boosting to control the complexity of tree leaf weights, reducing overfitting and improving robustness.