Neural Network Data Scrubbing for Privacy and Utility
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data anonymization methods face a trade-off between preserving analytically useful data and protecting privacy, as they struggle to minimize distortion while preventing sensitive information leakage.
Innovation Solution
The use of neural networks, specifically encoder, target task, bottleneck, and adversarial networks, to scrub images by preserving public attributes while removing private information, with a focus on minimizing the difference between estimated and preset attributes and the amount of information in the scrubbed feature, ensuring privacy protection and data utility.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data anonymization is performed to protect privacy, then privacy protection is improved, but analytically useful data is distorted
Solution Approach 1:
The patent segments data into public attributes and private attributes, processing them differently. Public attributes are preserved for analytical use while private attributes are anonymized, allowing simultaneous privacy protection and data utility maintenance through selective attribute handling
Solution Approach 2:
The patent introduces an intermediary anonymization layer that transforms original data into anonymized data through controlled generalization and suppression. This intermediary representation maintains analytical value while removing sensitive information, resolving the contradiction between privacy protection and data utility
2Productivity
If data is released to third parties for analysis, then data utility is improved, but information leakage risk increases
Solution Approach 1:
The patent extracts and removes sensitive private attributes from data before release to third parties, keeping only public attributes that are useful for analysis. This extraction process enables data sharing while eliminating information leakage risks associated with sensitive data
Solution Approach 2:
The patent changes data parameters by applying generalization (grouping specific values into broader categories) and suppression (removing or masking sensitive values). These parameter transformations reduce information leakage risk while preserving the analytical utility of the data for third-party analysis
Data Source
AI summary
An apparatus for scrubbing an image may include a memory storing instructions; and a processor configured to execute the instructions to: receive an input image; input a preset public attribute to an encoder neural network; obtain a scrubbed feature from the input image based on the preset public attribute, via the encoder neural network; wherein the encoder neural network is trained based on an amount of information in the scrubbed feature about the input image, and an estimated public attribute estimated from the scrubbed feature.