Quasi-Newton methods for machine learning: forget the past, just sample. (3rd September 2022)

• Record Type: Journal Article
• Authors: Berahas, A. S.; Jahani, M.; Richtárik, P.; Takáč, M.
• Is Part Of: Optimization methods and software. Volume 37:Number 5(2022)
• Page Start: 1668
• Page End: 1704
• Publication Date: 2022-09-03
• Subjects: Quasi-Newton -- curvature pairs -- sampling -- machine learning -- deep learning; Mathematical optimization -- Periodicals; Algorithms -- Periodicals; 519.7
• DOI: 10.1080/10556788.2021.1977806
• Languages: English
• ISSNs: 1055-6788

Abstract

Abstract : We present two sampled quasi-Newton methods (sampled LBFGS and sampled LSR1) for solving empirical risk minimization problems that arise in machine learning. Contrary to the classical variants of these methods that sequentially build Hessian or inverse Hessian approximations as the optimization progresses, our proposed methods sample points randomly around the current iterate at every iteration to produce these approximations. As a result, the approximations constructed make use of more reliable (recent and local) information and do not depend on past iterate information that could be significantly stale. Our proposed algorithms are efficient in terms of accessed data points (epochs) and have enough concurrency to take advantage of parallel/distributed computing environments. We provide convergence guarantees for our proposed methods. Numerical tests on a toy classification problem as well as on popular benchmarking binary classification and neural network training tasks reveal that the methods outperform their classical variants.