TabUnite: Efficient Encoding Schemes for Flow and Diffusion Tabular Generative Models

Jacob Si, Zijing Ou, Mike Qu, Yingzhen Li

Keywords: tabular data generation, encoding schemes, flow matching, diffusion

Abstract: Flow matching and diffusion generative models for tabular data face challenges in modeling heterogeneous feature interrelationships, especially in data with continuous and categorical input features. Capturing these interrelationships is crucial as it allows these models to understand complex patterns and dependencies in the underlying data. A promising option to address the challenge is to devise suitable encoding schemes for the input features before the generative modeling process. However, prior methods often rely on either suboptimal heuristics such as one-hot encoding of categorical features followed by separated modeling of categorical/continuous features, or latent space diffusion models. Instead, our proposed solution unifies the data space and jointly applies a single generative process across all the encodings, efficiently capturing heterogeneous feature interrelationships. Specifically, it employs encoding schemes such as PSK Encoding, Dictionary Encoding, and Analog Bits that effectively convert categorical features into continuous ones. Extensive experiments on datasets comprised of heterogeneous features demonstrate that our encoding schemes, combined with Flow Matching or Diffusion as our choice of generative model, significantly enhance model capabilities. Our TabUnite models help address data heterogeneity, achieving superior performance across a broad suite of datasets, baselines, and benchmarks while generating accurate, robust, and diverse tabular data.

Primary Area: generative models

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 5284