Secure Multi-Party Computation for Private ML Training and Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models trained on data from multiple sources face challenges in preserving data privacy and security, as sensitive information may leak when shared across different parties.
Innovation Solution
Implementing secure multi-party computation (MPC) techniques to train and perform inference operations using a neural network with a hidden layer for embedding input features, where each MPC computing system processes secret shares of data, ensuring that no single system can access the complete data in cleartext, and combining partial predictions to generate a final output.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data from multiple sources is shared across different parties for machine learning training, then computation efficiency and accuracy are improved, but data privacy and security deteriorate due to potential sensitive information leakage
Solution Approach 1:
The patent segments the machine learning model into multiple independent MPC computing systems, where each system holds only a portion (secret share) of the model parameters and data. This segmentation allows distributed computation across multiple parties while preventing any single party from accessing complete sensitive information, thus resolving the contradiction between computation efficiency and data privacy.
Solution Approach 2:
The patent introduces secure multi-party computation (MPC) protocols as an intermediary mechanism that enables collaborative machine learning training without direct data sharing. The MPC protocol acts as a mediator that allows multiple parties to jointly compute model updates while keeping their individual data and model parameters private, thereby maintaining both computation efficiency and data privacy.
2Object-affected harmful factors
If secret shares of data are processed by multiple MPC computing systems, then data privacy is enhanced, but device complexity increases due to the need for coordinated computation across multiple systems
Solution Approach 1:
The patent divides the complex machine learning training task into smaller sub-tasks that can be independently executed by each MPC computing system on their respective secret shares. This segmentation reduces the computational burden on each individual system while maintaining the overall privacy guarantees, thus managing device complexity effectively.
Solution Approach 2:
The patent combines the results from multiple independent MPC computing systems through secure aggregation protocols. Each system processes its local secret shares and contributes to the global model update, which is then aggregated to produce the final trained model. This merging approach distributes complexity across multiple systems while achieving the privacy benefits of secret share processing.
3Measurement precision
If a neural network with hidden layer for embedding is used, then computation accuracy is improved, but resource usage increases due to additional computational requirements
Solution Approach 1:
The patent segments the neural network architecture into distributed components across multiple MPC computing systems. The embedding layer and hidden layers are distributed such that each system processes only its portion of the computations on secret-shared data. This segmentation enables the use of accurate neural network architectures while reducing the resource burden on any single system through parallel distributed processing.
Data Source
AI summary
This disclosure relates to a privacy preserving machine learning platform. In one aspect, a method includes identifying a request for processing an input feature vector by a machine learning model using a multiple multi-party computation (MPC) cluster including a plurality of MPC computing systems. Each feature of the input feature vector is encoded to generate an encoded weight vector. A respective share of the encoded weight vectors is generated for each computing system and provided to a corresponding computing system to generate a partial prediction for the respective share. The MPC cluster collects modified partial predictions for the input feature vector from the rest of the multiple MPC computing systems, where each of the modified partial predications is generated based on the partial prediction by a corresponding MPC computing system. A final prediction is generated by the MPC cluster based on the respective partial predictions.