Privacy-Preserving ML Framework Using Homomorphic Encryption
A new framework enables training machine learning models on encrypted data without decryption, preserving privacy throughout the pipeline. The proof-of-concept uses Cheon-Kim-Kim-Song (CKKS) for approximate real-number arithmetic and demonstrates feasibility on K-Nearest Neighbors (KNN) and linear regression models, with encrypted inference for a basic Multilayer Perceptron (MLP) architecture. Published on arXiv (2604.23245), the work addresses privacy challenges in data-driven decision-making by allowing computations on encrypted data, preventing unauthorized access during processing.
Key facts
- arXiv paper 2604.23245 proposes privacy-preserving ML framework
- Uses Cheon-Kim-Kim-Song (CKKS) homomorphic encryption for approximate real-number arithmetic
- Demonstrates training KNN and linear regression models on encrypted data
- Evaluates encrypted inference for a basic Multilayer Perceptron (MLP) architecture
- Enables computations on encrypted data without decryption
- Addresses privacy challenges in data-driven decision-making
- Traditional encryption fails to secure data during processing
- Published as arXiv:2604.23245v1
Entities
Institutions
- arXiv