Authors:

Alan S Yi

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

The success of machine learning in healthcare relies on access to large, diverse datasets. Strict privacy and ethical regulations often hinder the ability to share data between different domains, such as manufacturing and drug design, which makes it difficult to train centralized machine learning models. To overcome this challenge, we propose decentralizing the data and training the models locally. This framework is known as federated learning (FL), where models are distributed and trained among multiple domains, and then each model is combined into a refined version (Dutta, 2024). In our most recent example of FL in hospital settings, each hospital trains models locally before their weights are sent to a server for aggregation between different clients (Bhatia, 2024). This research aims to build upon existing architecture by providing a federated learning framework for securing data between multiple clients through fully homomorphic encryption (FHE) and speeding up the process with quantum neural networks (QNNs). Enhancing federated learning with fully homomorphic encryption (FHE) ensures that even the model updates remain encrypted during transmission and aggregation, providing stronger end-to-end privacy protection (Bhatia & Bernal Neira, 2024). To simulate this process, we built a library that integrates the quantum-computing framework Pennylane (Bergholm et al., 2018) with the Pytorch-based federated-learning framework Flower (Beutel et al., 2020), all while utilizing fully homomorphic encryption to mask the data for further security. We benchmark FHE-QFL against centralized learning (CL) and traditional FL on various medical classification tasks, such as DNA and brain tumor MRI datasets. Our FHE-QFL models show AUC- ROC scores within 3% of traditional methods, demonstrating that quantum-enhanced federated models can enhance security while maintaining high accuracy. Keywords: Quantum Machine Learning; Federated Learning; Fully Homomorphic Encryption; Medical Classification † Presenting Undergrad Author; ‡ Contributing Undergrad Author; * Undergrad Acknowledgment

Source:

Purdue University / 2025

Topics:

No topics listed

Co-authors:

Alan S Yi