Publication

Medical AI, Monash University, Rethinking Semi-Supervised Federated Learning: A Dynamic Adaptive Threshold Generation Strategy

By

Sijin Zhou and Zongyuan Ge

January 12, 2024
Abstract

Researchers Sijin Zhou and Zongyuan Ge from Monash University leverage Rhino Federated Computing Platform (FCP) to power a multi-site consortium spanning Australia. Sijin presents his work in Federated Learning. In order to solve the problem of inflexibility, underutilization or over-utilization of data caused by the fixed threshold in semi-supervised Federated Learning, he proposes a framework named FedDAT, a Federated Dynamic Adaptive Threshold Generation Strategy. By dynamically updating the global and local thresholds on each unlabeled client, the framework is better able to capture the overall characteristics of the data without being overly affected by extreme values. They conducted experiments on a series of publicly available datasets, and the results demonstrate the superiority of our method. This method will be applied to real world datasets using Rhino Health's FCP and the Rhino Health Network.

Build your Federated Network