Secure Multi-Party Computation for ML Model Validation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for validating analytics or machine learning models are hindered by the need to share sensitive data or models, which can lead to intellectual property theft or data misuse.
Innovation Solution
The use of secure multi-party computation (SMPC) protocols allows for the validation of analytics or machine learning models by converting sensitive specifications into oblivious computations, compiling them into circuits compatible with SMPC protocols, and executing these computations securely without revealing sensitive inputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sensitive data or model specifications are shared for validation, then model performance can be evaluated, but intellectual property is exposed to theft or misuse
Solution Approach 1:
The patent introduces secure multi-party computation (SMPC) protocols as an intermediary mechanism that enables model validation without direct sharing of sensitive data or specifications. The SMPC protocol acts as a trusted mediator that allows the data owner to evaluate model performance on their private data while the model owner maintains confidentiality of their model specifications, thus resolving the contradiction between validation accuracy and IP protection
Solution Approach 2:
The patent segments the validation process into separate computational components that can be executed independently on encrypted data. By dividing the validation workflow into discrete SMPC-compatible operations, the system enables accurate model evaluation while keeping sensitive inputs segmented and protected throughout the computation process
2Object-affected harmful factors
If secure multi-party computation is used to protect sensitive information, then intellectual property confidentiality is maintained, but computational complexity increases
Solution Approach 1:
The patent creates simplified copies or representations of the model specifications that are compatible with SMPC protocols. Instead of working with complex original model representations, the system uses compiled circuit versions that maintain functional equivalence while being suitable for secure computation, thus reducing the complexity burden of IP protection
3Loss of information
If oblivious computations are compiled into SMPC circuits, then data confidentiality is maintained during validation, but processing time increases
Solution Approach 1:
The patent performs preliminary compilation of model specifications into SMPC-compatible circuit representations before the actual validation execution. This advance preparation work, including converting model logic into oblivious computations, is done once and can be reused for multiple validation runs, thereby amortizing the time cost of confidential processing across multiple uses
Data Source
AI summary
Systems and methods for validating analytics or machine learning models using secure multi-party computation (SMPC) are disclosed. The system operates with a data owner party providing a validation data set and an analytic owner party supplying a specification of the analytic or machine learning model. The analytic owner party converts the specification into an oblivious computation and compiles it into a circuit of operations compatible with an SMPC protocol. The data owner party downloads the compiled specification and constructs a validation circuit to execute the model against the validation data set without revealing the data or model details. The results of the SMPC computation are evaluated to produce performance statistics of the model, ensuring privacy for both parties' data throughout the process.