Secure Data Sharing with Multi-Key Homomorphic Encryption
Author Information
Author(s): Kang David Ha, Kim Duhyeong, Song Yongsoo, Lee Dongwon, Kwak Hyesun, Anthony Brian W.
Primary Institution: Massachusetts Institute of Technology
Hypothesis
Can a multi-party framework using multi-key homomorphic encryption enhance data privacy while allowing collaborative machine learning?
Conclusion
The MK-HELR algorithm enables secure and flexible collaborative learning by allowing new participants to join without disrupting the ongoing process.
Supporting Evidence
- The MK-HELR algorithm allows for dynamic membership, enabling new participants to join the learning process at any time.
- Models trained on aggregated data showed lower test errors compared to those trained on individual datasets.
- The algorithm maintains data privacy while facilitating collaborative learning among multiple parties.
Takeaway
This study shows how different companies can work together to learn from shared data without sharing their private information, using a special kind of encryption.
Methodology
The study developed the MK-HELR algorithm, which performs logistic regression on encrypted data from multiple contributors.
Limitations
The algorithm's performance may be constrained by the multiplicative depth and the number of participants involved.
Participant Demographics
The study utilized a predictive maintenance dataset with 10,000 data points, reflecting real-world manufacturing scenarios.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website