Heterogeneous Data-Aware Federated Learning for Medical AI

About this project

Federated learning is a technique that allows the decentralized training of machine learning models while preserving the user's privacy. As a result, federated learning brings the promise of providing a privacy-preserving mechanism for analyzing healthcare datasets. Yet, in their current implementations, federated learning algorithms assume independent and identically distributed (iiD) datasets. As a result, they might fail when facing real-world healthcare datasets where data is gathered from heterogeneous non-iiD sources. To address this issue, this project aims to investigate and develop strategies to deal with non-iiD data especially when applying federated learning to healthcare domains.