Privacy-Preserving ML Data Routing With Reversible Obfuscation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing machine learning systems face challenges in preserving data privacy by allowing only necessary data to be provided while maintaining the effectiveness of actionable inferences.

Innovation Solution

A method and system for obfuscating identifying information in data before transmission to machine learning providers and resolving it for actionable inferences, ensuring only non-identifying information is processed by the machine learning modules, with a reversible process to direct inferences to the correct endpoint.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If identifying information is provided to machine learning processes, then actionable inferences can be generated and directed to correct endpoints, but data privacy is compromised

Engineering Contradiction:
Improveactionable inference accuracyVSAvoiddata privacy risk
Core Design Contradiction:
ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The patent extracts identifying information from the data stream before processing, separating it from non-identifying information. Only non-identifying information is provided to machine learning processes, thereby protecting privacy while maintaining the ability to generate actionable inferences. The identifying information is retained separately for later matching purposes.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces an intermediary matching mechanism that connects machine learning outputs back to original data sources without exposing identifying information to the machine learning processes. This intermediary layer enables actionable inferences to be directed to correct endpoints while maintaining the privacy barrier.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Object-affected harmful factors

If only non-identifying information is provided to machine learning processes, then data privacy is preserved, but the ability to direct actionable inferences to correct endpoints is lost

Engineering Contradiction:
Improvedata privacy protectionVSAvoidendpoint identification capability
Core Design Contradiction:
Object-affected harmful factorsVSLoss of information

Solution Approach 1:

The patent segments the data processing into distinct stages: (1) extraction of identifying information before machine learning processing, (2) processing of only non-identifying information through machine learning, and (3) post-processing matching of results back to original endpoints. This segmentation preserves privacy at the processing stage while maintaining endpoint identification capability through separate matching operations.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary extraction and preservation of identifying information before the data is processed by machine learning algorithms. This preliminary action ensures that endpoint identification capability is maintained outside the machine learning process, allowing actionable inferences to be directed correctly after processing completes.

Inventive Principle:
Principle #10Preliminary action

3Productivity

If all data is collected for machine learning, then machine learning efficacy is improved, but data privacy concerns increase

Engineering Contradiction:
Improvemachine learning efficacyVSAvoiddata privacy concerns
Core Design Contradiction:
ProductivityVSObject-affected harmful factors

Solution Approach 1:

The patent extracts and removes identifying information from data before it is provided to machine learning processes. This extraction enables the collection and processing of large amounts of non-identifying information to improve machine learning efficacy, while the removed identifying information is protected from exposure to machine learning systems, thereby reducing privacy concerns.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP3971747B1Ambiguating and disambiguating data collected for machine learning
Publication Date: 2026.04.15 BLACKBERRY LTD
  • EP3971747B1 patent drawingFigure 1A
  • EP3971747B1 patent drawingFigure 1B
  • EP3971747B1 patent drawingFigure 2

AI summary

A method and privacy protection system for ambiguating and disambiguating data collected for machine learning. The method comprising receiving data from an endpoint, the data including identifying information for the endpoint and non-identifying information; obfuscating the identifying information in the received data to generate obfuscated data including the non-identifying information and obfuscated identifying information; transmitting the obfuscated data to one or more machine learning providers; receiving an actionable inference message based upon the non-identifying information in the obfuscated data from the one or more machine learning providers; resolving the obfuscated identifying information in the received actionable inference message to recover the identifying information; and transmitting the resolved actionable inference message to the endpoint associated with the identifying information. The privacy protection system comprising an endpoint, and obfuscator module, one or more machine learning modules, and a resolver module.