Bahrul Ilmi Nasution, a doctoral researcher at The University of Manchester, presented research on latent flow models for tabular data synthesis at the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases in Italy.
Study on Latent Flow Models
Bahrul Ilmi Nasution presented research titled Understanding Latent Flow Models for Tabular Data Synthesis: Targets, Paths, and Sampling at ECML PKDD 2026. Synthetic social data helps researchers access information derived from sensitive microdata without directly releasing original records.
The study investigated latent flow models, which are continuous-time generative models that transform random noise into synthetic observations. The research examined whether different learning approaches remain equivalent when implemented using finite data, approximate neural networks, numerical solvers, and limited computational budgets.
Performance Across Datasets
Across seven tabular datasets, including five censuses, the results showed that theoretically connected formulations can behave differently in practice. Velocity and posterior matching generally achieved higher statistical utility.
Meanwhile, score and noise matching tended to produce more conservative disclosure-risk profiles. The study also demonstrated that probability paths, sampling methods, numerical solvers, integration endpoints, and sampling budgets can substantially affect synthetic data quality.
Key facts and where they come from
- Bahrul Ilmi Nasution presented research titled Understanding Latent Flow Models for Tabular Data Synthesis: Targets, Paths, and Sampling at ECML PKDD 2026 in Italy.
Bahrul Ilmi Nasution, a doctoral researcher in the Department of Social Statistics at The University of Manchester, recently presented his research, Understanding Latent Flow Models for Tabular Data Synthesis: Targets, Paths, and Sampling, at the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD 2026) in Italy.
- Velocity and posterior matching generally achieve higher statistical utility.
Velocity and posterior matching generally achieve higher statistical utility
- Score and noise matching tend to produce more conservative disclosure-risk profiles.
score and noise matching tend to produce more conservative disclosure-risk profiles.
