Global Convergence of Block Coordinate Descent in Deep Learning

arXiv1803.00225MaRDI QIDQ6298409

Author name not available (Why is that?)

Publication date: 1 March 2018

Abstract: Deep learning has aroused extensive attention due to its great empirical success. The efficiency of the block coordinate descent (BCD) methods has been recently demonstrated in deep neural network (DNN) training. However, theoretical studies on their convergence properties are limited due to the highly nonconvex nature of DNN training. In this paper, we aim at providing a general methodology for provable convergence guarantees for this type of methods. In particular, for most of the commonly used DNN training models involving both two- and three-splitting schemes, we establish the global convergence to a critical point at a rate of $calO(1/k)$, where $k$ is the number of iterations. The results extend to general loss functions which have Lipschitz continuous gradients and deep residual networks (ResNets). Our key development adds several new elements to the Kurdyka-{L}ojasiewicz inequality framework that enables us to carry out the global convergence analysis of BCD in the general scenario of deep learning.

Has companion code repository: https://github.com/timlautk/BCD-for-DNNs-PyTorch

This page was built for publication: Global Convergence of Block Coordinate Descent in Deep Learning

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6298409)