Quasi-Newton Methods for Machine Learning: Forget the Past, Just Sample

Authors

Albert S. Berahas, University of Michigan, Ann Arbor
Majid Jahani, Lehigh University
Peter Richtárik, King Abdullah University of Science and TechnologyFollow
Martin Takáč, Mohamed Bin Zayed University of Artificial IntelligenceFollow

Document Type

Article

Publication Title

Optimization Methods and Software

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.

DOI

10.1080/10556788.2021.1977806

Publication Date

10-15-2021

Keywords

curvature pairs, deep learning, machine learning, Quasi-Newton, sampling

Comments

IR deposit conditions:

• OA - accepted version (pathway a)
• 12 month embargo
• Published source must be acknowledged
• Must link to publisher version
• Set statements to accompany deposits (see policy)

Recommended Citation

A. S. Berahas, M. Jahani, P. Richtárik, and M. Takáč, “Quasi-Newton methods for machine learning: forget the past, just sample,” Optimization Methods and Software, 2021, doi: 10.1080/10556788.2021.1977806.