The number of users affected by data breaches and cyberattacks in the United States has increased recently, making the country the most targeted in 2023. Concurrently, the evolution of the Internet of Things communications and machine learning algorithms has led to an unprecedented influx of data requiring analysis. This has intensified the urgency to identify efficient methods for safeguarding exchanged information during data communication and machine-learning training. Fully homomorphic encryption, operating on encrypted data without decryption, emerges as a promising solution. However, its practical integration faces challenges, notably in algorithm intricacy and computational constraints, especially regarding latency. This project aims to optimize resource-intensive operations in existing fully homomorphic encryption schemes and seamlessly integrate the optimized algorithm into federated learning frameworks. This will enhance security while preserving learning performance, revolutionizing secure data analysis in Internet of Things communications. By benefiting federated learning users and cloud providers, this project will enhance security in practical applications such as telehealth and wireless communications, contributing to enhanced privacy, security, and efficiency in critical sectors, ultimately advancing science for society.
This project develops an optimized fully homomorphic encryption algorithm to enhance security in Internet of Things (IoT) communications and integrates the optimized fully homomorphic encryption algorithm into federated learning frameworks to enable a secure training process on the encrypted data while maintaining learning performance. The project encompasses the following objectives: (1) developing a low-complexity, fully homomorphic encryption algorithm by optimizing resource-intensive operations; (2) integrating the optimized fully homomorphic encryption algorithm into federated learning to bolster security in IoT communications; and (3) accelerating fully homomorphic encryption and federated learning using parallel processing on graphics processing units and harnessing the combined computational power of both central processing units and graphics processing units. The proposed methodology is a pivotal stride toward propelling secure IoT communications into the future. By synergizing the domains of homomorphic encryption, parallel processing, and machine learning, this approach advances the field's theoretical underpinnings and introduces novel methodologies that contribute to its growth and evolution.
Quoc Bao Phan, Hien Nguyen, Phap Duong Ngoc, and Tuy Tan Nguyen, "Enhancing Data Security in Federated Learning with Dilithium," 43rd IEEE International Conference on Consumer Electronics (ICCE 2025), Las Vegas, NV, USA, 11–14 Jan. 2025, pp. 1–6. (Paper Link)(Source Code)
Quoc Bao Phan and Tuy Tan Nguyen, "Efficient Artificial Intelligence with Novel Matrix Transformations and Homomorphic Encryption," IEEE Journal on Emerging and Selected Topics in Circuits and Systems, vol. 14, no. 4, pp. 717–728, Dec. 2024. (Paper Link)(Source Code)
Linh Nguyen, Quoc Bao Phan, and Tuy Tan Nguyen, "Highly Reliable and Secure System with Multi-layer Parallel LDPC and Kyber for 5G Communications," IEEE Access, vol. 12, pp. 157260–157271, Oct. 2024. (Paper Link)(Source Code)
Quoc Bao Phan, Linh Nguyen, and Tuy Tan Nguyen, "Accelerating CKKS Homomorphic Encryption with Data Compression on GPUs," 67th International Midwest Symposium on Circuits and Systems (MWSCAS 2024), Springfield, MA, 11–14 Aug. 2024, pp. 1145–1149. (Paper Link)(Source Code)
Quoc Bao Phan, Linh Nguyen, Ngoc Thang Bui, Dinh C. Nguyen, Lan Zhang, and Tuy Tan Nguyen, "Federated Learning for Enhanced ECG Signal Classification with Privacy Awareness," 46th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC 2024), Orlando, Florida, USA, 15–19 Jul. 2024, pp. 1–4. (Paper Link)(Source Code)