Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Democratizing Tabular Data Access with an Open$\unicode{x2013}$Source Synthetic$\unicode{x2013}$Data SDK

Ivona Krchova, Mariana Vargas Vieyra, Mario Scriminaci, Andrey Sidorenko

TL;DR

The paper tackles the data bottleneck created by privacy and governance constraints by introducing an open-source, open-standards SDK for synthesizing high-quality tabular data. It centers on TabularARGN, an autoregressive generator capable of handling multi-table and sequential regimes, with optional differential privacy ($\epsilon$ budget) and fairness-aware generation, complemented by automated QA reports. The authors present a modular architecture (connectors, generators, synthetic datasets, QA) and a comprehensive workflow, plus support for LLM finetuning on text features and local deployment. Overall, the work demonstrates practical benefits for data democratization through privacy-preserving data synthesis, benchmarking performance against state-of-the-art methods and highlighting broad community adoption and real-world applicability.

Abstract

Machine learning development critically depends on access to high-quality data. However, increasing restrictions due to privacy, proprietary interests, and ethical concerns have created significant barriers to data accessibility. Synthetic data offers a viable solution by enabling safe, broad data usage without compromising sensitive information. This paper presents the MOSTLY AI Synthetic Data Software Development Kit (SDK), an open-source toolkit designed specifically for synthesizing high-quality tabular data. The SDK integrates robust features such as differential privacy guarantees, fairness-aware data generation, and automated quality assurance into a flexible and accessible Python interface. Leveraging the TabularARGN autoregressive framework, the SDK supports diverse data types and complex multi-table and sequential datasets, delivering competitive performance with notable improvements in speed and usability. Currently deployed both as a cloud service and locally installable software, the SDK has seen rapid adoption, highlighting its practicality in addressing real-world data bottlenecks and promoting widespread data democratization.

Democratizing Tabular Data Access with an Open$\unicode{x2013}$Source Synthetic$\unicode{x2013}$Data SDK

TL;DR

The paper tackles the data bottleneck created by privacy and governance constraints by introducing an open-source, open-standards SDK for synthesizing high-quality tabular data. It centers on TabularARGN, an autoregressive generator capable of handling multi-table and sequential regimes, with optional differential privacy ( budget) and fairness-aware generation, complemented by automated QA reports. The authors present a modular architecture (connectors, generators, synthetic datasets, QA) and a comprehensive workflow, plus support for LLM finetuning on text features and local deployment. Overall, the work demonstrates practical benefits for data democratization through privacy-preserving data synthesis, benchmarking performance against state-of-the-art methods and highlighting broad community adoption and real-world applicability.

Abstract

Machine learning development critically depends on access to high-quality data. However, increasing restrictions due to privacy, proprietary interests, and ethical concerns have created significant barriers to data accessibility. Synthetic data offers a viable solution by enabling safe, broad data usage without compromising sensitive information. This paper presents the MOSTLY AI Synthetic Data Software Development Kit (SDK), an open-source toolkit designed specifically for synthesizing high-quality tabular data. The SDK integrates robust features such as differential privacy guarantees, fairness-aware data generation, and automated quality assurance into a flexible and accessible Python interface. Leveraging the TabularARGN autoregressive framework, the SDK supports diverse data types and complex multi-table and sequential datasets, delivering competitive performance with notable improvements in speed and usability. Currently deployed both as a cloud service and locally installable software, the SDK has seen rapid adoption, highlighting its practicality in addressing real-world data bottlenecks and promoting widespread data democratization.

Paper Structure

This paper contains 31 sections, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Contributions
  3. Architecture Overview and Workflow
  4. Core Components of the SDK
  5. Connectors
  6. Generators
  7. Synthetic Datasets
  8. Synthetic Data Quality Assurance Report
  9. Basic workflow
  10. Data Ingestion and Processing
  11. Data Preparation
  12. Managing Multi-table Relationships
  13. Generative Model Training
  14. Auto-regressive Tabular Model Overview
  15. Support for LLM Finetuning
  16. ...and 16 more sections

Figures (6)

  • Figure 1:
  • Figure 2: Basic workflow example
  • Figure 3: Generator sharing
  • Figure 4: Training time (top) and accuracy (bottom) for the flat Adult (left) and sequential Baseball (right) datasets. Values are averaged over three or more runs; error bars indicate min/max. TabularARGN is shown both with and without DP training.
  • Figure 5: GitHub stars history for the SDK since its release on January 23rd 2025 until July 31st 2025.
  • ...and 1 more figures