Secure Multi-Party Computation for Privacy-Preserving Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning model training methods face challenges in ensuring data privacy and security, particularly in scenarios where multiple parties collaborate, as they are vulnerable to model inversion attacks and cannot guarantee the privacy of individual data contributions.
Innovation Solution
The implementation of a system that uses Secure Multi-Party Computation (SMPC) protocols combined with differential privacy (DP) techniques to train machine learning models. This system ensures data privacy by executing SMPC protocols on input data, applying DP techniques to the output, encrypting data before computations, and aggregating results from different parties to produce a collective output for model training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional machine learning training methods are used with multiple parties, then model training can be performed, but data privacy and security cannot be guaranteed and model inversion attacks are vulnerable
Solution Approach 1:
The patent segments the training data into shares distributed among multiple parties, where each party holds only a portion of the data in encrypted form. This segmentation enables collaborative training without any single party having access to complete sensitive data, thereby improving data privacy while maintaining training capability.
Solution Approach 2:
The patent introduces secure multi-party computation protocols as an intermediary mechanism that enables parties to jointly perform training operations on encrypted data without revealing their individual inputs. This intermediary layer ensures security and privacy while facilitating the training process.
2Reliability
If differential privacy techniques are applied to SMPC output, then individual data contributions remain private, but computational overhead increases
Solution Approach 1:
The patent applies differential privacy techniques to the output of the SMPC protocol rather than to the raw input data. This preliminary action on the output stage provides confidentiality guarantees while minimizing the impact on training efficiency, as the privacy protection is applied only to the final aggregated results.
3Reliability
If data is encrypted before computation in SMPC protocol, then data security is ensured, but computation speed decreases
Solution Approach 1:
The patent replaces traditional mechanical encryption methods with homomorphic encryption and secure multi-party computation protocols that enable computations to be performed directly on encrypted data. This substitution maintains data security throughout the computation process while reducing the overhead associated with repeated encryption and decryption operations.
Data Source
AI summary
Methods and systems for training machine learning models with enforced differential privacy using Secure Multi-Party Computation (SMPC) protocols are disclosed. A method for generating random bits for a Secure Multi-Party Computation (SMPC) protocol in a machine learning model training system includes locally generating, at each participating node, a plurality of random bits using a cryptographically secure pseudorandom number generator (CSPRNG). The locally generated random bits provided as input to the SMPC protocol and combined using an exclusive or (XOR) or addition operation to produce a set of combined random bits. A statistical sampling process for Differential Privacy (DP) is performed using the combined random bits as inputs. The perturbed training data is then utilized in the SMPC protocol to train the machine learning model.