Date

2020

Document Type

Thesis

Degree

Master of Science

Department

Industrial Engineering

First Adviser

Martin Takac

Abstract

In this thesis, we propose the Quantized SARAH (QSARAH) algorithm, which is a modified version of SARAH, that can be used in the federated learning setting. A practical variant of QSARAH, QSARAH+, is also proposed, which uses diminishing step sizes proper

Share

COinS