SUPR
Scalable Federated Learning
Dnr:

NAISS 2024/22-938

Type:

NAISS Small Compute

Principal Investigator:

Dejan Manojlo Kostic

Affiliation:

Kungliga Tekniska högskolan

Start Date:

2024-06-26

End Date:

2025-07-01

Primary Classification:

10201: Computer Sciences

Allocation

Abstract

The project "Scalable Federated Learning" aims to develop a highly scalable, flexible, and extensible distributed federated machine learning framework that can be applied to public health and wellness. This approach addresses significant limitations of existing federated learning systems, such as bandwidth requirements and inefficient handling of outliers. By leveraging advanced clustering, dropout techniques and anomaly detection techniques, this project seeks to enhance learning speed and accuracy from billions of devices, ultimately benefiting scenarios like privacy-preserving AI in healthcare and pandemic monitoring.