Inference-only method generates synthetic tabular temporal and relational data
Synthesis Without Training: An Inference-Only Pipeline for Tabular, Temporal, and Relational Synthetic Data
Machine Learning
Summary
Creating fake but realistic data usually requires training a computer model for each new dataset, which takes time and can be hard to understand. The authors introduce GENSCRIPT, a new approach that skips training altogether. Instead, it analyzes the original data to find patterns and rules, then uses a language model to write a program that can quickly produce similar synthetic data. This method works on different kinds of data, like tables, time series, and linked databases, and it produces samples nearly as accurate as existing methods while being faster and more transparent.
What this means in practice
- •For data engineers: Generate synthetic data for testing pipelines without costly model retraining or task-specific engineering.
- •For database administrators: Produce synthetic relational datasets preserving key and integrity constraints for development and backup validation.
Authors
Zilong Zhao, Abdul Raheem, Jiayu Li, Sohei Arisaka, Darius Lim Hong Yi, Milad Abdollahzadeh, Uzair Javaid, Biplab Sikdar
Abstract
Synthetic data generation is dominated by the fit-then-sample paradigm: a generative model is trained on a private dataset and then sampled from. Despite its widespread adoption, this paradigm faces three challenges: (1) a new training run is required for every dataset; (2) different data modalities, such as single tables, time series, and relational databases, require task-specific models and feature engineering; and (3) the resulting model is opaque, making its behavior under data constraints difficult to inspect. We propose GENSCRIPT, an inference-only pipeline that eliminates model training. GENSCRIPT computes a deterministic statistical profile of the source data (column types, ranges, missingness, categories, correlations, etc.) and passes it--rather than raw rows--to a language model to infer field semantics and cross-column integrity constraints. A coding agent then compiles the profile and constraints into an executable, auditable sampler. This unified approach supports single-table, temporal, and relational data without task-specific modeling. Across four single-table benchmarks, GENSCRIPT builds generators in 2 minutes and samples 50k rows within 6 seconds, while remaining within a few points of leading methods in marginal fidelity. Notably, it is the only method that perfectly preserves a 1-to-1 mapping between columns in the Adult dataset. On a smart-building dataset, it produces conditional time series that more closely match the real distribution than two baselines and perfectly preserves primary- and foreign-key relationships in the corresponding relational database.