GAN Synthetic Data Generation with Multi-Discriminator Privacy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating synthetic data using generative adversarial networks (GANs) face a trade-off between privacy guarantees and performance, as differential privacy (DP) can degrade the performance of machine learning models, necessitating a method to maintain model performance while ensuring DP.
Innovation Solution
A synthetic data generation apparatus employing three discriminators - a first discriminator to distinguish actual and synthetic data, a second discriminator to ensure differential privacy, and a third discriminator to refine synthetic data generation, using algorithms like DP-GAN and PATE-GAN, allowing sequential learning of the generator to produce data indistinguishable from actual data while satisfying DP.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If differential privacy is applied to protect privacy in synthetic data generation, then privacy protection is improved, but machine learning model performance deteriorates
Solution Approach 1:
The patent divides the single discriminator into three specialized discriminators: a first discriminator for general data distribution discrimination, a second discriminator for differential privacy verification, and a third discriminator for synthetic data quality assessment. This segmentation allows each discriminator to focus on specific aspects, enabling the system to maintain model performance while ensuring differential privacy protection in the synthetic data generation process.
2Device complexity
If a single discriminator is used in GAN to simplify the system, then device complexity is reduced, but the ability to simultaneously ensure privacy and performance is insufficient
Solution Approach 1:
The patent segments the discrimination function into three specialized discriminators, each handling specific aspects of the generation process. This segmentation resolves the contradiction by demonstrating that while device complexity increases, the system gains the capability to simultaneously ensure both differential privacy and model performance, which a single discriminator cannot achieve.
Solution Approach 2:
The generator serves multiple functions by being trained with three different discriminators simultaneously. It must satisfy the data distribution requirements of the first discriminator, the differential privacy constraints of the second discriminator, and the quality standards of the third discriminator. This multi-functionality approach allows a single generator to achieve multiple objectives that would otherwise require separate systems.
Data Source
AI summary
A synthetic data generation apparatus according to an embodiment includes a generator for generating synthetic data from an input value, a first discriminator learned to distinguish between actual data and the synthetic data, a second discriminator learned to distinguish between the actual data and the synthetic data while satisfying differential privacy, and a third discriminator learned to distinguish between first synthetic data which is output from the generator learned by the first discriminator and second synthetic data which is output from the generator learned by the second discriminator.