RTDL (Research on Tabular Deep Learning)

RTDL (Research on Tabular Deep Learning) is a collection of papers and packages on deep learning for tabular data.

:bell: To follow announcements on new projects, subscribe to releases in this GitHub repository: "Watch -> Custom -> Releases".

[!NOTE] The list of projects below is up-to-date, but the rtdl Python package is deprecated. If you used the rtdl package, please, read the details.

1. First, to clarify, this repository is NOT deprecated, only the package rtdl is deprecated: it is replaced with other packages.
2. If you used the latest rtdl==0.0.13 installed from PyPI (not from GitHub!) as pip install rtdl, then the same models (MLP, ResNet, FT-Transformer) can be found in the rtdl_revisiting_models package, though API is slightly different.
3. :exclamation: If you used the unfinished code from the main branch, it is highly recommended to switch to the new packages. In particular, the unfinished implementation of embeddings for continuous features contained many unresolved issues (the rtdl_num_embeddings package, in turn, is more efficient and correct).

Papers

(2026) Benchmarking Optimizers for MLPs in Tabular Deep Learning

Paper   Code

(2025) Unveiling the Role of Data Uncertainty in Tabular Deep Learning

Paper

(2025) On Finetuning Tabular Foundation Models

Paper   Code

(2024) TabM: Advancing Tabular Deep Learning with Parameter-Efficient Ensembling

Paper   Code   Usage

(2024) TabReD: Analyzing Pitfalls and Filling the Gaps in Tabular Deep Learning Benchmarks

Paper   Code

(2023) TabR: Tabular Deep Learning Meets Nearest Neighbors

Paper   Code

(2022) TabDDPM: Modelling Tabular Data with Diffusion Models

Paper   Code

(2022) Revisiting Pretraining Objectives for Tabular Deep Learning

Paper   Code

(2022) On Embeddings for Numerical Features in Tabular Deep Learning

Paper   Code   Package (rtdl_num_embeddings)

(2021) Revisiting Deep Learning Models for Tabular Data

Paper   Code   Package (rtdl_revisiting_models)

(2019) Neural Oblivious Decision Ensembles for Deep Learning on Tabular Data

Paper   Code