Secure Multi-Party Computation for Privacy-Preserving Customer Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Business entities face limitations in sharing customer data due to privacy regulations, hindering their ability to enrich customer data for informed business decisions while ensuring customer privacy.
Innovation Solution
A system and method utilizing secure multi-party computation to generate an artificial intelligence model from shared customer data schemas, allowing each business entity to predict customer classifications without exchanging actual data, using a predictor manager to distribute the model for execution on their respective customer data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If business entities share customer data to enrich their customer data information base, then their ability to make informed business decisions is improved, but customer privacy is compromised and privacy regulations are violated
Solution Approach 1:
A privacy-preserving computation platform acts as an intermediary between business entities, enabling them to compute customer classifications using their respective customer data without the data itself being shared or exposed to other entities. The platform provides secure multi-party computation capabilities that allow collaborative analysis while maintaining data privacy.
Solution Approach 2:
The system segments the customer data processing into separate, isolated computations that each business entity performs on their own data. Instead of sharing complete customer data, each entity contributes only necessary data schemas and receives aggregated classification results, preventing privacy violations while enabling collaborative insights.
2Object-affected harmful factors
If business entities are limited in sharing customer data due to privacy regulations, then customer privacy is protected, but their ability to enrich customer data for business decisions is hindered
Solution Approach 1:
The privacy-preserving computation platform serves as a mediator that enables data-rich computations without data sharing. Business entities can access enriched customer information through the platform's secure computation capabilities, which aggregate insights from multiple entities while keeping individual data private and compliant with privacy regulations.
Solution Approach 2:
The system changes the fundamental parameter of data sharing from direct data exchange to computation result sharing. Instead of sharing raw customer data, entities share processed classification predictions and aggregated statistics, transforming the nature of information exchange to comply with privacy regulations while maintaining analytical capabilities.
3Measurement precision
If business entities exchange actual customer data to generate predictive models, then model accuracy is improved, but data security and privacy compliance are compromised
Solution Approach 1:
The platform acts as a secure intermediary that enables collaborative model training without direct data exchange. Each business entity contributes their local customer data and schemas to the platform, which coordinates secure multi-party computations to generate unified predictive models. The platform ensures data remains encrypted and inaccessible to other entities throughout the process.
Solution Approach 2:
The model training process is segmented into distributed computations where each entity processes their own data independently under the coordination of the platform. Classification models are trained on segmented data portions and then aggregated, ensuring high prediction accuracy through comprehensive data utilization while maintaining data security through isolation and encryption.
Data Source
AI summary
As described herein, a system, method, and computer program are provided for using shared customer data and artificial intelligence to predict customer classifications. A first system of a first business entity receives an artificial intelligence model generated using output of a secure multi-party computation applied to: a first schema of first customer data stored by the first system, and a second schema of second customer data stored by a second system of a second business entity. Additionally, the first system executes the artificial intelligence model on the first customer data stored by the first system to generate a predictor, the predictor configured to receive input and process the input to predict a classification for the input. Further, the first system distributes the predictor for use by the second system of the second business entity to predict at least one classification for the second customer data.