Research map / Image generation / Autoregressive and latent generation
Q-Tab: Quantized Tabular Data Generator
Julian Wustl, Philipp Haid, Yarema Okhrin, Claudius Schnörr
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add residual LFQ with residual-only bit-flip corruption instead of flat LFQ with global corruption (ablation) | Q-Tab tokenizer · Shoppers/Beijing/Default/Adult mean | MLE gap (%): 11.24 -> 7.80 | Improved |
| add bit-flip corruption to flat LFQ (ablation) | LFQ tokenizer · mean over 4 datasets | duplicate rate (%): 30.54 -> 9.60 | Improved |
| add global bit-flip corruption to flat LFQ (ablation) | LFQ tokenizer · mean over 4 datasets | MLE gap (%): 7.99 -> 11.24 | Hurt |
| use residual corruption versus none (no corruption LFQ) (ablation) | Q-Tab · mean over 4 datasets | MLE gap (%): 7.99 -> 7.80 | No clear effect |
| sample categorical tokens first (Cat->Num) instead of numerical first (ablation) | Q-Tab sampler · Shoppers, Beijing | DCR: 0.6176 -> 0.534 | Improved |
| sample categorical tokens first instead of uniform random order (ablation) | Q-Tab sampler · Shoppers, Beijing | quality: 0.979 -> 0.982 | Improved |
| change sampling order (ablation) | Q-Tab sampler · Shoppers, Beijing | MLE (AUC/RMSE): AUC 0.919 / 0.9176 / 0.919 | No clear effect |
| increase LFQ bit size B beyond 3-4 (ablation) | Q-Tab · sensitivity analysis | utility vs duplicates | Mixed |
| increase bit-flip corruption epsilon above 0.05 (ablation) | Q-Tab · sensitivity analysis | AUCROC | Hurt |
| add moderate MLM label smoothing on residual tokens (ablation) | Q-Tab · sensitivity analysis | weighted-F1 | Improved |
| use Q-Tab instead of TabDiff for MLE | Q-Tab · Average gap to real data | average gap (%): 6.0 -> 4.6 | Improved |
| use Q-Tab instead of TabDiff for distributional fidelity | Q-Tab · Average DF gap | DF gap (%): 2.0 -> 2.2 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.
- TabDiff — TabDiff: a Mixed-type Diffusion Model for Tabular Data Generation (ICLR 2025) · better on 4 of 8 shared results
- TabNat — TabNAT: A Continuous-Discrete Joint Generative Framework for Tabular Data (ICML 2025) · better on 1 of 1 shared result
- TabDDPM — TabDDPM: Modelling Tabular Data with Diffusion Models (ICML 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- TabNAT: A Continuous-Discrete Joint Generative Framework for Tabular Data (ICML 2025 · similarity 0.87)
- Large Scale Transfer Learning for Tabular Data via Language Modeling (NeurIPS 2024 · similarity 0.86)
- TRL-Bench: Standardizing Cross-Paradigm Representation-Level Evaluation of Tabular Encoders (NeurIPS 2026 · similarity 0.85)
- TabSDS: a Lightweight, Fully Non-Parametric, and Model Free Approach for Generating Synthetic Tabular Data (ICML 2025 · similarity 0.85)
- TabM: Advancing tabular deep learning with parameter-efficient ensembling (ICLR 2025 · similarity 0.85)
- Multimodal Tabular Reasoning with Privileged Structured Information (NeurIPS 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
