Core Concepts
Utilizing Full Homomorphic Encryption for Improved Security, Efficiency, and Practicality in Federated Learning.
Abstract
The article discusses the advancements in Federated Learning Schemes using Full Homomorphic Encryption (FHE). It introduces novel schemes that enhance security, functionality, and practicality. By comparing different datasets from various fields like medical, business, biometric, and financial sectors, the study shows significant improvements in security, efficiency, and practicality over traditional federated learning schemes. The use of FHE algorithms allows for direct additions or multiplications on ciphertexts, enhancing the redesign of federated learning frameworks. Additionally, the article highlights the importance of security against quantum computing attacks and gradient attacks provided by FHE algorithms. The efficiency of FHE algorithms has significantly improved in recent years, making them more practical for training modules compared to traditional schemes using Partial Homomorphic Encryption (PHE).
Stats
Comparisons have been given in four practical data sets separately from medical, business, biometric and financial fields.
Our scheme has achieved a great efficiency improvement in training modules compared with classical schemes using PHE algorithms.
The horizontal logistic regression federated learning model is 9.3 times more efficient than the classical algorithm for training.
The vertical logistic regression federated learning model is 3-3.7 times more efficient than the classical algorithm for training.
Quotes
"Our scheme achieves significant improvements in security, efficiency and practicality."
"The FHE algorithm can provide controllable security for models and data."
"Our proposed federated learning scheme based on FHE achieves unification in the training process."