Accelerated Gradient Boosting
This is an incremental improvement for machine learning practitioners using gradient boosting, as it enhances efficiency and robustness in prediction tasks.
The authors tackled the problem of gradient boosting's sensitivity to hyperparameters and model complexity by combining it with Nesterov's accelerated descent, resulting in AGB, which is less sensitive to shrinkage and produces sparser models while maintaining high performance.
Gradient tree boosting is a prediction algorithm that sequentially produces a model in the form of linear combinations of decision trees, by solving an infinite-dimensional optimization problem. We combine gradient boosting and Nesterov's accelerated descent to design a new algorithm, which we call AGB (for Accelerated Gradient Boosting). Substantial numerical evidence is provided on both synthetic and real-life data sets to assess the excellent performance of the method in a large variety of prediction problems. It is empirically shown that AGB is much less sensitive to the shrinkage parameter and outputs predictors that are considerably more sparse in the number of trees, while retaining the exceptional performance of gradient boosting.