SUPR
Scalable Federated Learning For Privacy-Preserving Training of Machine Learning Models
Dnr:

NAISS 2024/22-1358

Type:

NAISS Small Compute

Principal Investigator:

Li Ju

Affiliation:

Uppsala universitet

Start Date:

2024-11-01

End Date:

2025-11-01

Primary Classification:

10105: Computational Mathematics

Webpage:

Allocation

Abstract

The main aim of the project is to develop new algorithms and tools for federated machine learning (FL). FL is a recent development that addresses both privacy issues with traditional centralized machine learning, as well as practical issues related to large or fast data near the computational edge (decentralized AI). From a scientific computing standpoint, FL is related to distributed and decentralized optimization, however an important difference from traditional HPC approaches is that we do not have any control over the partitioning of data on different compute nodes or devices. This, along with the considerable system heterogeneity of real world applications, introduces unique challenges when it comes to efficiency and convergence.The project aims at efficient and robust federated training schemes, implementation, experiments and software.