Barzilai–Borwein-based adaptive learning rate for deep learning

•A Barzilai–Borwein-based method for adaptive learning rate of training DNNs.•The method is highly insensitive to initial learning rate which greatly reduces computational effort.•The method has its advantage over others on learning speed and generalization performance.•Convergence guarantee of the...

Ausführliche Beschreibung

Bibliographische Detailangaben

Veröffentlicht in:	Pattern recognition letters 2019-12, Vol.128, p.197-203
Hauptverfasser:	Liang, Jinxiu, Xu, Yong, Bao, Chenglong, Quan, Yuhui, Ji, Hui
Format:	Artikel
Sprache:	eng
Schlagworte:	Adaptive algorithms Adaptive learning rate Barzilai–Borwein method Deep learning Deep neural network Learning curves Methods Motivation Neural networks Stochastic gradient descent Training
Online-Zugang:	Volltext
Tags:	Tag hinzufügen Keine Tags, Fügen Sie den ersten Tag hinzu!