April 2023 Summaries

Gretel Tabular DP is a new model that generates high quality tabular synthetic data with mathematical guarantees of privacy. It's a differentially private graph-based generative model that creates synthetic versions of sensitive data, offering provable mathematical guarantees of privacy. The model works well on datasets with primarily categorical variables, relatively low cardinality (<100 unique categories per variable) and under 100 variables. It follows the select-measure-generate paradigm developed by McKenna et al., which involves selecting a subset of correlated pairs of variables using a differentially private algorithm, measuring distributions of the selected pairs with differential privacy, and estimating a probabilistic graphical model that captures the relationship as described by the noisy marginals.

Gretel Tabular DP is a new model for generating high quality tabular synthetic data with mathematical guarantees of privacy. It is a differentially private graph-based generative model that creates synthetic versions of sensitive data, offering provable mathematical privacy guarantees. The model works well on datasets with primarily categorical variables and under 100 variables. Gretel Tabular DP can produce synthetic records maintaining high statistical symmetry with the original data in just 10 minutes, even with very conservative differential privacy budgets (ε < 1, δ < 10^-7).

Gretel has introduced a new tool, Gretel Evaluate, which allows users to assess the performance of synthetic data on machine learning models with just a few clicks. This feature provides a Synthetic Data Utility Report and an ML Quality Score (MQS), making it easier for organizations to validate the utility of their synthetic data. The tool can be used to synthesize and evaluate data, customize models, and generate reports comparing synthetic data performance with real-world data.