Synthetic Data Testing for Machine Learning Compliance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge lies in determining whether synthetic data is sufficiently similar to original, protected data to be used effectively in machine learning applications, while ensuring compliance with varying data protection regulations across different jurisdictions.
Innovation Solution
A method involving a computing device that accesses a protected batch of data and generates a simulated batch of anonymized data. By comparing variables between the protected and simulated data, a similarity value is obtained, and if it exceeds a threshold, the computing device performs machine learning functions using the simulated data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If synthetic data is generated from protected data, then data protection compliance is improved, but data quality and similarity to original data deteriorates
Solution Approach 1:
The patent generates synthetic data by copying statistical properties and patterns from protected data while anonymizing individual records. The synthetic data replicates key characteristics such as distributions, correlations, and patterns without containing actual personal information, thus maintaining compliance while preserving data utility for machine learning.
Solution Approach 2:
The patent transforms protected data into synthetic data by changing the state of individual records while preserving statistical parameters. The synthetic data maintains similar statistical properties (means, variances, correlations) but changes the specific values and patterns at the record level, achieving both compliance and data quality.
2Reliability
If synthetic data is used in machine learning models, then data protection compliance is improved, but model accuracy deteriorates
Solution Approach 1:
The patent implements a feedback mechanism where the synthetic data is generated, then compared against the protected data using statistical tests. The results of this comparison feed back into the data generation process, allowing iterative refinement to ensure the synthetic data maintains sufficient similarity for accurate machine learning models while remaining compliant.
Solution Approach 2:
The patent performs preliminary statistical analysis and characterization of the protected data before generating synthetic data. This preliminary action includes computing statistical summaries, distributions, and patterns that will be used to guide the synthetic data generation, ensuring the output will maintain the necessary quality for accurate machine learning.
3Reliability
If data is anonymized to protect privacy, then data protection compliance is improved, but data utility for machine learning deteriorates
Solution Approach 1:
The patent applies different levels of anonymization and synthetic data generation to different aspects of the data. Critical identifying information is fully anonymized or synthesized, while statistical properties and patterns are preserved. This local quality approach ensures compliance for sensitive fields while maintaining utility for machine learning in other fields.
Solution Approach 2:
The patent creates a composite data structure that combines anonymized/protected fields with synthetic fields. The resulting synthetic dataset maintains the structural and statistical properties needed for machine learning while replacing sensitive information with anonymized or generated data, achieving both compliance and utility.
Data Source
AI summary
Determining whether synthetic data is sufficient for utilization in connection with one or more machine learning models. The computing device accesses a protected batch of data associated with a machine learning model. The computing device accesses a simulated batch of data, the simulated batch of data based upon but anonymizing the protected batch of data. The computing device accesses one or more comparisons of one or more variables in the protected batch of data and the simulated batch of data to obtain a similarity value. The computing device performs a machine learning function utilizing at least in-part the simulated batch of data if the similarity value exceeds a similarity threshold.