Neural Network Data Obfuscation for Privacy Protection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional methods for anonymizing data in big data fail to accurately conceal personal identification information without damaging the original data, and the concealed information can often be reversed, making it difficult to protect privacy and hinder the big data trade market.

Innovation Solution

A method and device that use a learning network to obfuscate data through network operations and backpropagation, generating irreversibly anonymized data that is recognizable as similar to the original data by computers but distinct to humans, by updating data using characteristic information and task-specific outputs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional anonymization techniques (masking or blurring) are applied to conceal identification information, then personal information protection is improved, but the original data is damaged and machine learning cannot utilize the information

Engineering Contradiction:
Improvepersonal information protectionVSAvoidoriginal data integrity
Core Design Contradiction:
ReliabilityVSLoss of information

Solution Approach 1:

The patent applies parameter changes by transforming the original data through learned transformation parameters that obfuscate identification information while preserving task-relevant features. The transformation modifies data parameters (such as pixel values, feature representations) in a controlled manner that prevents identification but maintains utility for machine learning tasks.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent introduces an intermediary obfuscation model that acts as a mediator between the original data and the anonymized output. This intermediary transforms the data through learned representations, allowing indirect access to information while preventing direct identification. The intermediary structure enables privacy protection without complete information loss.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If identification information is masked or blurred in the data, then personal information protection is improved, but machine learning cannot utilize the information due to damage to original data

Engineering Contradiction:
Improvepersonal information protectionVSAvoidmachine learning utility
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent applies local quality by differentiating the treatment of different parts of the data. Identification information regions are transformed to prevent recognition, while other regions retain more original characteristics to preserve machine learning utility. The obfuscation model learns to apply different transformation strengths to different local regions based on their importance for identification versus task performance.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent segments the data into identification-critical regions and task-relevant regions, applying different transformation strategies to each segment. This segmentation allows the system to protect privacy in identification regions while maintaining utility in task-relevant regions, resolving the contradiction between protection and usability.

Inventive Principle:
Principle #1Segmentation

3Measurement precision

If conventional video anonymization blurs all pixels with changes between frames, then motion detection is improved, but critical information such as facial expression becomes distorted and personal identification information may remain

Engineering Contradiction:
Improvemotion detection accuracyVSAvoidfacial expression information
Core Design Contradiction:
Measurement precisionVSLoss of information

Solution Approach 1:

The patent applies dynamics by using temporal information and motion patterns to guide the obfuscation process. Instead of static blurring, the system dynamically adjusts transformation based on frame differences and motion characteristics, preserving essential dynamic information while preventing identification. The obfuscation adapts to temporal changes rather than applying uniform masking.

Inventive Principle:
Principle #15Dynamics

4Reliability

If blurring is applied to anonymize video images, then personal information protection is improved, but the blurred video image can be reverted to the original using deblurring techniques

Engineering Contradiction:
Improvepersonal information protectionVSAvoidanonymization irreversibility
Core Design Contradiction:
ReliabilityVSStability of the object's composition

Solution Approach 1:

The patent replaces mechanical blurring operations with learned transformation mechanisms based on neural networks. Instead of simple spatial averaging that can be reversed through deconvolution, the system uses complex non-linear transformations that are computationally infeasible to invert. The substitution of mechanical operations with learned models creates irreversible obfuscation.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP3827365B1Method for concealing data and data obfuscation device using the same
Publication Date: 2024.12.25 DEEPING SOURCE INC
  • EP3827365B1 patent drawingFigure 1
  • EP3827365B1 patent drawingFigure 2
  • EP3827365B1 patent drawingFigure 3

AI summary

A method for concealing original data to protect personal information is provided. The method includes steps of: a data obfuscation device (a) if the original data is acquired, inputting the original data or its modified data into a learning network, and allowing the learning network to (i) apply a network operation to the original data or the modified data using learned parameters of the learning network and thus to (ii) output characteristic information on the original data or the modified data; and (b) updating the original data or the modified data via backpropagation using part of (i) 1-st losses calculated by referring to the characteristic information and its corresponding 1-st ground truth, and (ii) 2-nd losses calculated by referring to (ii-1) a task specific output generated by using the characteristic information and (ii-2) a 2-nd ground truth corresponding to the task specific output, to thereby generate obfuscated data.