Synthetic Document Generation Using Kernel Density Estimates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating synthetic documents often result in inaccuracies, which in turn lead to reduced accuracy in machine learning models trained on these documents, due to the need to avoid private or sensitive information.
Innovation Solution
The use of kernel density estimates (KDEs) generated from original documents to create highly accurate synthetic documents that do not include private information, while also being computationally efficient.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If current methods are used to generate synthetic documents while avoiding private information, then data privacy is maintained, but accuracy of the synthetic documents and machine learning models is reduced
Solution Approach 1:
The patent introduces kernel density estimates (KDEs) as an intermediary representation between original documents and synthetic documents. The KDE captures the statistical distribution of the original data without containing actual private information, serving as a privacy-preserving mediator that still enables accurate synthetic document generation.
Solution Approach 2:
The patent transforms the original documents into a different parameter space by computing kernel density estimates. This parameter transformation allows the synthetic documents to be generated from statistical properties rather than direct copying, maintaining accuracy while ensuring privacy through mathematical abstraction.
2Reliability
If traditional synthetic document generation methods are used, then data privacy is protected, but computing resources required are high
Solution Approach 1:
The patent uses kernel density estimates as a compact mathematical representation (copy) of the original data distribution. Instead of storing or processing multiple copies of actual private documents, the system works with this compressed statistical representation, significantly reducing computing resources while maintaining privacy.
Data Source
AI summary
A service may generate keywords to produce synthetic documents, while maintaining data privacy for the original documents. A client may extract keyword sequences from locally stored documents, embed the keyword sequences into vectors, and generate a DP-KDE distribution based on the vectors. The DP-KDE distribution preserves data privacy for the original documents. The service receives the DP-KDE distribution, obtains a particular vector from the DP-KDE (e.g., based on a calculated score for the DP-KDE using random Gaussian completions), decodes the particular vector into a sequence of synthetic keywords, and uses the sequence of synthetic keywords to prompt an LLM to produce one or more synthetic documents.