Privacy-Preserving EEG Data Generation: A Federated Split Learning Approach Using Privacy-Adaptive Autoencoders and Secure Aggregation with GFlowNet

Shouvik Paul, Garima Bajwa

2025

Abstract

EEG-based Brain-Machine Interfaces (BMI) are novel interaction paradigms used extensively in assistive technologies and neurorehabilitation. However, these interfaces pose significant privacy risks as they rely on unique neural patterns for their operation, which unintentionally reveal sensitive cognitive information and biometric identifiers without consent. Unlike traditional data, EEG signals are challenging to anonymize due to their complex, high-dimensional, and noise-sensitive nature. We present a novel approach to privacy-preserving EEG data generation, combining Federated Split Learning (FSL) with hierarchical privacy-adaptive autoencoders, secure aggregation, and Generative Flow Networks (GFlowNet). The hierarchical architecture of the autoencoder enables multi-level feature extraction, effectively capturing both spatial and temporal de-pendencies in the EEG signals. Using Rényi Differential Privacy (RDP) and adaptive noise scaling, our model anonymizes sensitive brain signals during data generation. The FSL architecture allows client-side processing of raw EEG data, followed by server-side reconstruction and synthetic data generation using GFlowNet. Secure aggregation further enhances privacy, ensuring that individual data contributions are protected even during client and server communication. Evaluations of our approach under various privacy budgets demonstrate a balanced privacy-utility trade-off.

Paper Citation

in Harvard Style

Paul S. and Bajwa G. (2025). Privacy-Preserving EEG Data Generation: A Federated Split Learning Approach Using Privacy-Adaptive Autoencoders and Secure Aggregation with GFlowNet. In Proceedings of the 22nd International Conference on Security and Cryptography - Volume 1: SECRYPT; ISBN 978-989-758-760-3, SciTePress, pages 638-643. DOI: 10.5220/0013529100003979

in Bibtex Style

@conference{secrypt25,
author={Shouvik Paul and Garima Bajwa},
title={Privacy-Preserving EEG Data Generation: A Federated Split Learning Approach Using Privacy-Adaptive Autoencoders and Secure Aggregation with GFlowNet},
booktitle={Proceedings of the 22nd International Conference on Security and Cryptography - Volume 1: SECRYPT},
year={2025},
pages={638-643},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013529100003979},
isbn={978-989-758-760-3},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 22nd International Conference on Security and Cryptography - Volume 1: SECRYPT
TI - Privacy-Preserving EEG Data Generation: A Federated Split Learning Approach Using Privacy-Adaptive Autoencoders and Secure Aggregation with GFlowNet
SN - 978-989-758-760-3
AU - Paul S.
AU - Bajwa G.
PY - 2025
SP - 638
EP - 643
DO - 10.5220/0013529100003979
PB - SciTePress