Synthetic Tabular Data Generation via Variational Autoencoder and GAN
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing synthetic data generation techniques for tabular data fail to adequately account for inter-feature correlation distributions, making the generated data unsuitable for training machine learning models, and they also pose privacy risks due to the inclusion of personal information.
Innovation Solution
A machine learning model based on a variational autoencoder and Generative Adversarial Network (GAN) framework is used to generate synthetic tabular data by learning inter-feature correlations from real data sources, introducing noise to mask features, and iteratively inferring their values, thus preserving the correlations in the generated data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If real data is collected from real world sources to train machine learning models, then the quality and authenticity of training data is improved, but privacy issues arise due to personal information being traceable back to real people
Solution Approach 1:
The patent creates synthetic copies of real tabular data through a two-stage machine learning process. First, a variational autoencoder learns the underlying distribution and inter-feature correlations of real data. Then, a generative adversarial network uses this learned distribution to generate synthetic data records that replicate the statistical properties and correlations of real data without containing actual personal information, thus providing high-quality training data while eliminating privacy risks
Solution Approach 2:
The patent transforms real data into synthetic data by changing the fundamental parameter of data origin from real-world collection to algorithmic generation. The variational autoencoder learns the probability distribution parameters of real data, and the generative model samples from this distribution to create synthetic records with matching statistical characteristics but different underlying values, effectively decoupling data quality from privacy concerns
2Object-affected harmful factors
If existing synthetic data generation techniques are used, then privacy concerns are addressed, but inter-feature correlation distributions are not adequately preserved making the data unsuitable for training machine learning models
Solution Approach 1:
The patent introduces a variational autoencoder as an intermediary between the real data and the generative adversarial network. This intermediary component explicitly learns and captures the inter-feature correlation distributions of real data by training on the complete tabular dataset. The learned correlations are then transferred to the GAN through the encoded latent representation, ensuring that synthetic data generated by the GAN preserves realistic feature relationships while maintaining privacy protection
Solution Approach 2:
The patent performs preliminary learning of inter-feature correlations using the variational autoencoder before the actual synthetic data generation process. The VAE is trained first to understand the correlation structure of real data, encoding this knowledge into its latent space. Only after this preliminary correlation learning is complete does the GAN begin generating synthetic data, ensuring that correlation preservation is built into the generation process from the start rather than being an afterthought
3Reliability
If more real data is collected to improve model training, then model performance is improved, but the challenges of collecting and storing personal information increase
Solution Approach 1:
Instead of collecting and storing additional real personal information data, the patent creates synthetic copies that replicate the statistical properties and inter-feature correlations of real data. These synthetic data records serve as substitutes for real data in the training process, providing the same model training benefits without requiring expanded data collection and storage infrastructure for personal information
Solution Approach 2:
The patent extracts only the essential statistical properties and inter-feature correlation structures from real data through the variational autoencoder's learning process. This extracted knowledge is then used by the generative model to create synthetic data, separating the useful training characteristics from the problematic personal information, thereby eliminating the need for extensive real data collection and storage infrastructure
Data Source
AI summary
Systems and methods for generating synthetic tabular data for machine learning and other applications are provided. In some embodiments, a variational autoencoder is trained to learn inter-feature correlations found in tabular data collected from real data sources. The trained variational autoencoder is used to train a generator model of a Generative Adversarial Network (GAN) to generate synthetic tabular data that exhibits the inter-feature correlation distribution found in the tabular data collected from real data sources. In some embodiments, processing devices perform operations comprising: receiving a set of tabular data records, each record comprising a plurality of features; training a first machine learning model using the tabular data records to learn correlations between the plurality of features; and training a second machine learning model, using the first machine learning model, to generate a synthetic tabular data records based at least on the one or more correlations between the plurality of features.