MOSAIC-FL, a micro-service based privacy-preserving framework with application to genomics

2026-07-27

AI Summary

This work addresses the challenges of security, privacy, robustness, and communication efficiency in federated learning within sensitive domains such as healthcare and genomics by proposing a microservices-based privacy-preserving federated learning framework. The framework employs a modular architecture, leveraging gRPC for communication and finite state machines to synchronize components and detect threats. It innovatively integrates threshold CKKS homomorphic encryption, dynamic collective key updates, and noise flooding techniques to achieve IND-CPA-D security while effectively mitigating key recovery attacks. Empirical evaluations on EMNIST image classification and TCGA breast cancer subtype classification tasks demonstrate the framework’s high efficiency, scalability, and adaptability across varying model sizes and threshold configurations.

Abstract

Security and privacy are primordial requirements for Federated Learning (FL), especially in fields such as healthcare and genomics where sensitive information has to be analyzed. Our FL framework is designed to address these challenges while proposing a modular, flexible and micro-service architecture. More precisely, it integrates an efficient gRPC communication layer and a Finite State Machine to ensure robust component synchronization and threat detection, while relying on a fault-tolerant secure aggregation protocol using a Threshold variant of the CKKS homomorphic cryptosystem. This allows blind model aggregation by an orchestration server, requiring a minimum of $t$-out-of-$N$ active clients for decryption while minimizing communication overhead thanks to both cryptographic and network protocols. We ensure IND-CPA-D security through noise flooding and mitigate the recent key-recovery attack on synchronized decryptors by renewing the collective key material at every round. We demonstrate the framework's effectiveness through diverse use cases, ranging from standard image recognition (EMNIST) to complex genomic classification including breast cancer subtyping on TCGA, evaluating system performance across different threshold values and model scales.

Problem

• Federated Learning
• Privacy Preservation
• Genomics
• Secure Aggregation
• Homomorphic Encryption

Innovation

• Threshold CKKS
• secure aggregation
• micro-service architecture
• federated learning
• privacy-preserving genomics