Current browse context:
stat.ML
Change to browse by:
References & Citations
Statistics > Machine Learning
Title: Infinitesimal gradient boosting
(Submitted on 26 Apr 2021 (v1), last revised 24 Jan 2023 (this version, v2))
Abstract: We define infinitesimal gradient boosting as a limit of the popular tree-based gradient boosting algorithm from machine learning. The limit is considered in the vanishing-learning-rate asymptotic, that is when the learning rate tends to zero and the number of gradient trees is rescaled accordingly. For this purpose, we introduce a new class of randomized regression trees bridging totally randomized trees and Extra Trees and using a softmax distribution for binary splitting. Our main result is the convergence of the associated stochastic algorithm and the characterization of the limiting procedure as the unique solution of a nonlinear ordinary differential equation in a infinite dimensional function space. Infinitesimal gradient boosting defines a smooth path in the space of continuous functions along which the training error decreases, the residuals remain centered and the total variation is well controlled.
Submission history
From: Jean-Jil Duchamps [view email][v1] Mon, 26 Apr 2021 15:09:05 GMT (40kb)
[v2] Tue, 24 Jan 2023 20:05:55 GMT (113kb,D)
Link back to: arXiv, form interface, contact.