Token-Based Communication for Privacy-Preserving Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in sharing customer behavioral knowledge derived from machine learning without violating individual privacy or exposing proprietary business practices, as personally identifiable information and sensitive data are subject to regulatory restrictions.
Innovation Solution
The use of tokens representing private data events or groups, which are used to train machine learning systems without revealing the underlying information, allowing organizations to share knowledge without exposing protected data, and enabling secure exchange of customer behavioral insights.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If personally identifiable information is shared between entities for machine learning training, then the quality and accuracy of machine learning models improve, but regulatory restrictions and privacy violations occur
Solution Approach 1:
The patent introduces tokens as intermediary elements that represent personally identifiable information without containing the actual sensitive data. These tokens serve as mediators between the need for accurate model training and privacy protection requirements, allowing machines to process and learn from data patterns while preventing direct exposure of individual identities. The tokenization process creates a layer of abstraction where the intermediary token system enables information exchange without compromising privacy.
2Adaptability or versatility
If proprietary business information is shared for collaborative machine learning, then the versatility and applicability of models improve, but exposure of trade secrets and proprietary practices occurs
Solution Approach 1:
The patent extracts and separates the essential learning signals from proprietary business information through tokenization. By taking out only the necessary patterns and relationships needed for model training while leaving behind sensitive proprietary details, the system enables collaborative learning without exposing trade secrets. This extraction process allows organizations to contribute to collective model improvement while retaining control over their proprietary information.
3Reliability
If raw private data is used for training machine learning systems, then the reliability of training data improves, but security risks and compliance issues increase
Solution Approach 1:
The patent creates token copies that represent private data without being the actual sensitive information. These token copies maintain the structural and relational properties necessary for reliable machine learning training while being safe to share and process. The copying mechanism allows the system to work with representations of data that preserve training quality without introducing the security risks associated with handling raw private information.
Data Source
AI summary
Techniques are described for communicating between two organizations without exchanging sensitive private information. One of the methods includes generating a token representative of private data. The method includes identifying at least one entity associated with the private data. The method includes associating the token with at least one entity. The method also includes providing information identifying at least one entity and the token to a machine learning system.