Core Concepts
Proposing a secure and efficient distributed machine learning scheme based on MKTFHE.
Abstract
The content introduces a novel distributed decryption protocol for MKTFHE, implements privacy-preserving logistic regression and neural networks, and proposes a secure framework for privacy-preserving machine learning. The implementation details, accuracy, efficiency analysis, security considerations, and experimental results are discussed comprehensively.
Content Structure:
Introduction to Distributed Machine Learning with Privacy Concerns
Abstract - Addressing Security Risks in Decryption of MKTFHE
Proposed Attack on Existing Decryption Protocol & Introduction of Secret Sharing Scheme
Design of New Activation Function for MKTFHE-Friendly Operations
Implementation of Logistic Regression & Neural Networks with Privacy Preservation
Framework for Secure Distributed Machine Learning
Implementation & Experiment Details - Accuracy & Efficiency Analysis
Conclusion & Acknowledgment
Stats
b = 1/4m - Σ⟨ai, si⟩ + e (mod 1)
Efficiency of our function is 10 times higher than using 7-order Taylor polynomials directly.
Accuracy achieved is similar to using high-order polynomial activation function schemes.
Quotes
"We develop a secure distributed decryption protocol for MKTFHE by introducing a secret sharing scheme."
"Our proposed activation function can shorten the computing activation function time in ciphertext by 10 times."