Efficient Privacy-Preserving K-Means Clustering from Secret-Sharing-Based Secure Three-Party Computation.

Area of Science:

• Computer Science
• Cryptography
• Machine Learning

Background:

• Privacy concerns necessitate privacy-preserving machine learning (PPML).
• Existing PPML methods face efficiency challenges compared to plain-text algorithms.
• K-means clustering is a fundamental data mining technique often requiring sensitive data.

Purpose of the Study:

• To design a novel, efficient, and privacy-preserving K-means clustering algorithm.
• To address the efficiency gap in current privacy-preserving machine learning techniques.
• To leverage secret sharing for secure outsourced computation in clustering tasks.

Main Methods:

• Developed a K-means clustering algorithm based on replicated secret sharing.
• Implemented the algorithm within the semi-honest model using three computing servers.
• Utilized secret sharing to enable secure outsourcing of the clustering task.

Main Results:

• The proposed privacy-preserving scheme guarantees full data privacy.
• Experimental results show the privacy-preserving K-means achieves accuracy comparable to plain-text versions.
• Achieved significant performance improvements: 16.5×-25.2× faster computation and 63.8×-68.0× lower communication compared to existing schemes.

Conclusions:

• The novel secret-sharing-based K-means algorithm offers a practical solution for privacy-preserving data clustering.
• The scheme demonstrates high efficiency and accuracy, making it suitable for secure outsourced computation.
• This work advances the field of privacy-preserving machine learning with a focus on practical efficiency.