Related Experiment Video
Updated: Sep 6, 2025
08:45
Mapping Cortical Dynamics Using Simultaneous MEG/EEG and Anatomically-constrained Minimum-norm Estimates: an Auditory Attention Example
Published on: October 24, 2012
14.8K
A Correlated Noise-assisted Decentralized Differentially Private Estimation Protocol, and its application to fMRI
Hafiz Imtiaz1, Jafar Mohammadi2, Rogers Silva3
1Department of Electrical and Electronic Engineering, Bangladesh University of Engineering and Technology, Dhaka, Bangladesh.
IEEE Transactions on Signal Processing : a Publication of the IEEE Signal Processing Society
|June 27, 2022
Summary
We developed a differentially private algorithm for decentralized independent component analysis (ICA) in neuroimaging. Jointly distributed noise enhances privacy-utility trade-offs, matching non-private utility in some cases.
Area of Science:
- Neuroimaging Analysis
- Computational Neuroscience
- Privacy-Preserving Machine Learning
Background:
- Independent Component Analysis (ICA) is crucial for neuroimaging data analysis.
- Decentralized learning is needed for large, privacy-sensitive datasets.
- Centralized analysis is often impossible due to data privacy concerns.
Purpose of the Study:
- To propose a differentially private algorithm for decentralized ICA.
- To address the utility limitations of conventional private decentralized algorithms.
- To investigate the benefits of limited collaboration via correlated noise.
Main Methods:
- Developed a novel differentially private algorithm for decentralized ICA.
- Introduced the use of jointly distributed random noise to improve privacy-utility.
- Validated the approach using synthetic and real neuroimaging data.
Main Results:
- The proposed algorithm achieves differential privacy in a decentralized setting.
- Jointly distributed noise significantly improves the privacy-utility trade-off.
- Comparable utility to non-private methods was achieved for specific parameters.
Conclusions:
- Meaningful utility is achievable in complex signal processing systems while preserving privacy.
- Decentralized differentially private ICA is feasible for neuroimaging.
- Limited collaboration through correlated noise is a promising strategy.
