Pith. sign in

REVIEW 5 major objections 5 minor 34 references

Dependency-aware synthetic tabular data generation

T0 review · 5 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A two-step generation framework preserves inter-attribute rules in synthetic tabular data.

desk verdict A sensible wrapper for enforcing known FD/LD rules into synthetic tabular data, but the benchmark evaluation is circular and the 'consistent improvement' claim goes beyond the evidence. read the letter →

arxiv 2507.19211 v1 pith:Y7YM2HZR submitted 2025-07-25 cs.LG

Chaithra Umesh , Kristian Schultz , Manjunath Mahendra , Saptarshi Bej , Olaf Wolkenhauer This is my paper · ORCID
classification cs.LG
keywords synthetictabulardatafunctionaldependencieslogicalhierarchicalfeaturegenerationgenerativemodelsCTGANTVAEGReaT
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that synthetic tabular data can preserve inter-attribute rules—functional dependencies (FDs) and logical dependencies (LDs)—much more reliably if the generative model is not asked to reproduce the full joint distribution. The proposed Hierarchical Feature Generation Framework (HFGF) generates only the independent features with a standard generative model, then reconstructs the dependent features from predefined mapping rules. The authors evaluate this on four controlled benchmark datasets with known dependencies, using six generative models, and report that the wrapper improves dependency preservation in nearly all settings. This matters because privacy-sensitive fields such as healthcare need synthetic copies that obey the same deterministic and rule-based constraints as the original records.

What carries the argument

The load-bearing object is a configuration dictionary that encodes each feature's type and, for every dependent feature, a mapping rule from the parent feature's categories to the dependent feature's categories; one-to-one and many-to-one mappings are deterministic functional dependencies, while one-to-many mappings are probabilistic logical dependencies. Algorithm 1 builds the benchmark datasets from this dictionary, and the same dictionary drives the reconstruction of dependent features after the generative model produces synthetic independent features. The framework's effect is measured by FDTool, which mines functional dependencies, and the Q-function, a score between 0 and 1 in which 0 marks functional dependence, 1 marks independence, and intermediate values mark logical dependence.

What would settle it

Run HFGF on a benchmark in which one independent categorical feature has a rare class that the base generative model never emits; if the preserved FD and LD percentages then fall to roughly the no-HFGF baseline, the improvement is conditional on class coverage rather than intrinsic to the framework.

Watch

Extended reading notes

Core claim

The central claim is that dependency preservation is a separation problem rather than a modeling problem: once the features of a table are split into an independent set and a dependent set, any generative model trained only on the independent features, combined with a reconstruction step that applies the known mapping rules, produces synthetic tables in which the known FDs and LDs survive far more often than when the model generates every column jointly. Preservation is measured as the percentage of benchmark dependencies that FDTool and the Q-function rediscover in the synthetic table. In the paper's four benchmark scenarios, every tested model—CTGAN, CTABGAN+, TVAE, NextConvGeN, TabuLa, and GReaT—preserves more FDs with HFGF than without, and in the complex imbalanced case all six also preserve more LDs; two small-case comparisons are called out where TVAE preserves more LDs and NextConvGeN preserves more FDs without the wrapper.

Load-bearing premise

The dependency structure must be known in advance and expressible as categorical mapping rules, and the base generative model must produce every category value of the independent features that those mappings require.

Editorial extensions

If this is right

  • A user who knows the dependency structure of a table can preserve that structure without modifying the generative model itself, simply by wrapping it in HFGF.
  • Models that normally require large datasets can keep dependencies intact on tables as small as 100 rows, because the generative step only has to learn a low-dimensional distribution of a few independent features.
  • Dependency preservation becomes a checkable engineering condition: if every category of each independent feature appears in the synthetic sample, the reconstructed dependent features are correct by construction.
  • The framework applies only when the table has genuinely independent features and the dependencies are categorical mapping rules; fully mutually dependent tables fall outside its scope.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural next test is to apply HFGF to real-world datasets whose dependencies are inferred by FDTool rather than predefined; the gain would then depend on the completeness and accuracy of those inferred rules.
  • Because the bottleneck shifts to the independent features, an obvious practical add-on is a coverage check that resamples or conditions until every class needed by the mappings appears.
  • The same separation idea could extend to numerical-numerical and numerical-categorical dependencies by first discretizing numerical features into bins, an extension the authors name as future work.
  • The harder problem the framework side-steps is learning the dependency structure itself; a model that inferred FDs and LDs from real data would remove the need for a hand-written configuration dictionary.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper proposes the Hierarchical Feature Generation Framework (HFGF), which separates features into independent and dependent sets based on known functional and logical dependencies, generates the independent features with a standard generative model, and then reconstructs the dependent features by applying the predefined mapping rules. The authors create four synthetic benchmark datasets with controlled dependency structure and evaluate six generative models (CTGAN, CTABGAN+, TVAE, NextConvGeN, TabuLa, GReaT) with and without HFGF, reporting the percentage of FDs and LDs preserved as measured by FDTool and a Q-function. The stated conclusion is that HFGF consistently improves FD/LD preservation across all six models.

Significance. If the claim held as stated, the framework would offer a simple, model-agnostic way to enforce known categorical dependencies in synthetic tabular data, which is relevant for privacy-sensitive domains such as healthcare. The paper also contributes controlled benchmark datasets with known dependencies and makes code available. However, the central evaluation is largely circular: the dependent features in the benchmark are generated from the same mapping rules that HFGF uses for reconstruction, and the evaluation tools detect exactly those rules. The experiments therefore do not test whether generative models can learn dependencies, but only whether the independent features are generated with sufficient fidelity. This substantially limits the significance of the empirical results, although the benchmark resource and the explicit separation of independent and dependent features may still be useful to the community.

major comments (5)
  1. [Section 3, Algorithm 1 and Section 4, Tables 1-3] The central evaluation is circular. In Algorithm 1, the benchmark datasets are constructed by setting dependent features using fi.mapping rules (one-to-one, many-to-one, one-to-many). HFGF then reconstructs dependent features in the synthetic data by applying exactly those same mapping rules to the generated independent features, and preservation is measured with FDTool and the Q-function, which detect exactly those rules. Consequently, the high FD/LD preservation scores under HFGF are a definitional consequence of the construction rather than an empirical finding about the generative models' ability to preserve dependencies. The results in Tables 1-3 essentially test only whether the generative model reproduces the needed categories of the few independent features. To make the claim meaningful, the authors should either explicitly frame HFGF as an oracle-based post-processing wrapper for known dependencies or add experiments that break this circularity, for example by deriving dependencies from a real dataset with FDTool and then applying HFGF, or by comparing against a base generative model that attempts to model all features jointly while evaluated on the same dependencies.
  2. [Abstract, Introduction, and Section 4, Tables 1-3] The claim that HFGF "consistently improves" preservation is contradicted by the paper's own tables. In Table 1, TVAE LD preservation drops from 60 to 39 (Case 1); in Table 2, TVAE LD preservation drops from 71 to 33 (Case 3); in Table 3, NextConvGeN FD preservation drops from 100 to 75 (Case 1) and from 100 to 86 (Case 3). While Section 4 acknowledges two of these decreases, the abstract, introduction, and conclusion repeat the unqualified claim. The manuscript should state the actual scope, e.g., "improves in most cases" or "improves on average", and should analyze why the decreases occur, since the text currently attributes only the TVAE LD drops to mode collapse and does not reconcile these counterexamples with the headline claim.
  3. [Section 4, Tables 1-3] All reported preservation percentages are point estimates from a single run per model and case, with no error bars, standard deviations, or number of seeds. Since the six generative models are stochastic and two of the datasets have only 100 rows, the observed jumps (e.g., CTGAN FD preservation from 0 to 75) and the decreases (e.g., NextConvGeN from 100 to 75) could be dominated by seed variance. The authors should report results over multiple random seeds (at least five, ideally ten) with mean and standard deviation, or justify why the effect sizes are robust to seed choice. Without this, the quantitative claims in Tables 1-3 are not statistically grounded.
  4. [Section 4 and Figure 2] The manuscript does not specify how "preserved LDs (%)" is computed from the Q-function scores. The Q-function returns a continuous value in [0,1], where 0 indicates a functional dependency and intermediate values indicate logical dependency, but the text does not define the threshold or matching rule that converts scores into a binary "preserved" decision for the LD set. Similarly, it is unclear how the set of LDs in the benchmark data is enumerated from Q-scores and how the comparison with synthetic data is performed. Please provide an explicit operational definition of LD preservation, as this is essential for interpreting Tables 1-3.
  5. [Section 5, Limitations and Future Directions] The framework assumes that the dependency structure is known in advance and expressible as categorical mapping rules, and the authors state in the Failure cases paragraph that "only four independent features are generated, which proves to be tractable for the models even with limited training data." This indicates that the evaluated scenario is low-dimensional and that the generative task reduces to modeling a few uncorrelated marginals. The paper should test the approach in a more demanding setting, for instance with many independent features (more than ten) or with numerical dependent features, to assess scalability and to support the claim that HFGF can be applied generally in clinical settings. As written, the limitations section acknowledges the constraint but the conclusions still generalize beyond the evidence.
minor comments (5)
  1. [Title page and CRediT statement] The author names are inconsistent between the title page ("Chaithra Umesha", "Kristian Schultza") and the CRediT statement ("Chaithra Umesh", "Kristian Schultz"); please ensure the names match throughout.
  2. [Equation (1)] The Q-function definition in Equation (1) has garbled typesetting: the fraction and the condition are compressed into one line. Please rewrite with proper display format and clearly define the symbols A, B, T, and the relation ~T.
  3. [Figure 2 caption] The caption states "Incorporation of HFGF consistently improves dependency preservation across all models," but Tables 1-3 contain counterexamples (NextConvGeN Cases 1 and 3; TVAE Cases 1 and 3). The caption should be revised to reflect the actual, qualified results.
  4. [Section 5, Figure 3] The claim that "models with HFGF not only improve dependency preservation but also align better with real data distributions" is not fully supported by the Peacock test p-values: for TVAE the p-value decreases from 0.077 without HFGF to 0.017 with HFGF, and for CTGAN the increase is from 0.006 to 0.063, which is borderline. Please provide a more systematic comparison of the p-values or a formal statistical test over multiple embeddings.
  5. [Section 1, one-to-one FDs] The statement that in one-to-one FDs "either feature may be treated as independent" is useful but should be justified or accompanied by a short example, since it affects the determination of the independent set in Algorithm 1.

Circularity Check

3 steps flagged · score 8.0 of 10

The claimed improvement is a definitional consequence: dependent features are constructed from the same mapping rules used to build the benchmarks and then detected by FDTool and the Q-function.

  1. self definitional [Section 3 (Algorithm 1 and HFGF step 2); Section 3.1]
    "Dependent features are reconstructed by applying the mapping rules specified in the configuration dictionary. These mappings are specified explicitly for each dependent feature as shown in the second step of Figure 1."

    Algorithm 1 constructs every benchmark dependent feature as fi.mapping[category] (or a probabilistic one-to-many sample) from its parent independent feature. HFGF's second step applies exactly the same configuration mappings to the synthetic independent features. Therefore any synthetic independent sample that retains the parent's categories yields dependent features that satisfy the benchmark's FDs and LDs by construction; no dependency is learned or inferred. The subsequent 'improvement' is a property of the wrapper, not an empirical capability of the generative models.

  2. self definitional [Section 4, Figure 2 caption; Section 3 third step]
    "Preservation is computed as the percentage of benchmark data FDs /LDs that are also discovered in the synthetic data using FDTool and the Q-function."

    FDTool and the Q-function are used to measure whether the synthetic data contains the very dependencies that were inserted through fi.mapping when the benchmark and the synthetic dependent features were created. Because dependent features are defined by those mappings, the FDs and LDs are recovered whenever the independent-feature categories are present. The preservation score therefore measures a definitional consequence of the construction rather than an independent prediction. The authors' Q-function [4] is part of the same measurement loop, adding self-citation but not changing the reduction.

1 more flagged steps
  1. other [Section 6, Conclusion]
    "It is essential to acknowledge that the current iteration of HFGF primarily acts as a framework to identify and map existing dependencies from real data onto synthetic data, rather than inherently learning these complex relationships within the generative model itself."

    This statement concedes that HFGF does not learn dependencies; it maps pre-specified rules. Together with Algorithm 1's use of the same fi.mapping to build the benchmarks and to reconstruct synthetic dependent features, the admission confirms that the reported improvements are a definitional consequence of injecting oracle rules, not an empirical discovery. It also limits the framework to known categorical dependencies (stated as a limitation at the end of Section 5), so the central claim does not generalize to dependency discovery or numerical interactions.

full rationale

HFGF's central result — that wrapping a generative model with HFGF improves FD/LD preservation — holds by construction, not by empirical discovery. Algorithm 1 creates benchmark dependent features via fi.mapping[category], and Section 3 states that HFGF reconstructs dependent features by applying the same mapping rules. The evaluation then uses FDTool and the Q-function to check for exactly those dependencies, so the preservation percentages in Tables 1–3 are forced as long as the generated independent features cover the needed categories. The paper itself narrows the empirical remainder to class coverage: 'Since both FDs and LDs are functions of the independent features, their preservation depends on the accurate generation of those independent variables' and notes in the Conclusion that HFGF 'primarily acts as a framework to identify and map existing dependencies ... rather than inherently learning these complex relationships.' There is genuine residual content — whether base models reproduce the categories of the few independent features — and the framework is transparent about requiring known categorical rules, but the headline claim 'HFGF improves the preservation of FDs and LDs' is tautological in this benchmark design. Score 8 rather than 10 because the category-coverage experiments are not entirely vacuous and the framework is explicitly described as a mapping wrapper; however, the central quantitative evidence reduces by construction.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The framework rests on the premise that all dependencies can be expressed as deterministic or probabilistic mappings between categorical features, that these mappings are known in advance (via config or FDTool inference), and that the base generative model adequately covers the independent feature space. These are domain assumptions, not standard mathematical axioms.

assumptions (4)
  • domain assumption The dependency structure is known in advance and expressible as acyclic categorical mapping rules.
    Algorithm 1 requires a config dictionary with predefined mappings for each dependent feature; the framework cannot infer or learn these rules on its own.
  • domain assumption FDTool and the Q-function correctly identify all FDs and LDs in both benchmark and synthetic data.
    These tools are used as ground truth for evaluation in Section 3, step 3, but their own accuracy is not independently validated in the paper.
  • domain assumption The base generative model, when trained only on independent features, will produce every category needed to apply the mapping rules.
    Section 5 'Failure cases' states that if a generative model misses a class in an independent feature, dependent features relying on it lose the intended dependency.
  • domain assumption Inter-attribute dependencies can be fully captured by categorical feature mappings.
    Section 5 'Limitations' acknowledges that the current setup considers only dependencies among categorical features and does not model numerical-categorical interactions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dependency-aware synthetic tabular data generation." pith.science (2026). https://pith.science/paper/Y7YM2HZR

@misc{pith2026250719211,
  author       = {Pith},
  title        = {Pith review of: Dependency-aware synthetic tabular data generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Y7YM2HZR}},
  note         = {Machine review of arXiv:2507.19211}
}
read the original abstract

Synthetic tabular data is increasingly used in privacy-sensitive domains such as health care, but existing generative models often fail to preserve inter-attribute relationships. In particular, functional dependencies (FDs) and logical dependencies (LDs), which capture deterministic and rule-based associations between features, are rarely or often poorly retained in synthetic datasets. To address this research gap, we propose the Hierarchical Feature Generation Framework (HFGF) for synthetic tabular data generation. We created benchmark datasets with known dependencies to evaluate our proposed HFGF. The framework first generates independent features using any standard generative model, and then reconstructs dependent features based on predefined FD and LD rules. Our experiments on four benchmark datasets with varying sizes, feature imbalance, and dependency complexity demonstrate that HFGF improves the preservation of FDs and LDs across six generative models, including CTGAN, TVAE, and GReaT. Our findings demonstrate that HFGF can significantly enhance the structural fidelity and downstream utility of synthetic tabular data.

Figures

Figures reproduced from arXiv: 2507.19211 by the authors.

Figure 1
Figure 1. Hierarchical feature generation framework to improve the preservation of FDs and LDs in synthetic data. The workflow includes: (1) generating benchmark datasets under controlled conditions, (2) using generative models to synthesize independent features and mapping dependent features with known dependencies, and (3) evaluating dependency preservation using FDTool and Q-function on both benchmark and synthetic dataset… view at source ↗
Figure 2
Figure 2. Radar chart comparing the preservation of dependencies with and without Hierarchical Feature Generation Framework (HFGF) using six generative models. Preservation is computed as the percentage of benchmark data FDs/LDs that are also discovered in the synthetic data using FDTool and the Q-function. The evaluation is based on the fourth benchmark dataset, which includes 40 FDs, 324 LDs, and exhibits feature imbalance.… view at source ↗
Figure 3
Figure 3. 2-dimensional PCA embeddings of real data and synthetic data generated by six different generative models, with and without the use of HFGF, on fourth benchmark data. For each synthetic dataset, the corresponding p-value (shown above each plot) is computed using the Peacock test, which evaluates whether the real and synthetic embeddings are drawn from the same distribution. A p-value Greater than 0.05 suggests no st… view at source ↗

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 17 canonical work pages

  1. [1]

    T. Lee, An Information-Theoretic Analysis of Relational Databases—Part I: Data Dependencies and Information Metric, IEEE Transactions on Software Engineer- ing SE-13 (10) (1987) 1049–1061. doi:10.1109/TSE.1987.232847

    work page arXiv 1987

  2. [2]

    J. Liu, J. Li, C. Liu, Y . Chen, Discover dependencies from data—a review, IEEE Transactions on Knowledge and Data Engineering 24 (2) (2012) 251–264. doi: 10.1109/TKDE.2010.197

    work page doi:10.1109/tkde.2010.197 2012

  3. [3]

    Xu, C.-T

    S. Xu, C.-T. Lee, M. Sharma, R. B. Yousuf, N. Muralidhar, N. Ramakrishnan, Are LLMs Naturally Good at Synthetic Tabular Data Generation? (Jun. 2024). doi:10.48550/arXiv.2406.14541

    arXiv doi:10.48550/arxiv.2406.14541 2024

  4. [4]

    Umesh, K

    C. Umesh, K. Schultz, M. Mahendra, S. Bej, O. Wolkenhauer, Preserving logical and functional dependencies in synthetic tabular data, Pattern Recognition 163 (2025) 111459. doi:10.1016/j.patcog.2025.111459

    work page arXiv 2025

  5. [5]

    Y . Long, L. Xu, A. Brintrup, LLM-TabFlow: Synthetic Tabular Data Gener- ation with Inter-column Logical Relationship Preservation (Mar. 2025). doi: 10.48550/arXiv.2503.02161

    arXiv doi:10.48550/arxiv.2503.02161 2025

  6. [6]

    Buranosky, E

    M. Buranosky, E. Stellnberger, E. Pfa ff, D. Diaz-Sanchez, C. Ward-Caviness, FDTool: a Python application to mine for functional dependencies and can- didate keys in tabular data, F1000Research 7 (2019) 1667. doi:10.12688/ f1000research.16483.2

    work page 2019

  7. [7]

    Papenbrock, F

    T. Papenbrock, F. Naumann, A Hybrid Approach to Functional Dependency Dis- covery, in: Proceedings of the 2016 International Conference on Management 19 of Data, SIGMOD ’16, Association for Computing Machinery, New York, NY , USA, 2016, pp. 821–833. doi:10.1145/2882903.2915203

    arXiv 2016

  8. [8]

    H. Yao, H. J. Hamilton, Mining functional dependencies from data, Data Mining and Knowledge Discovery 16 (2) (2008) 197–219. doi:10.1007/ s10618-007-0083-9

    work page 2008

Show all 34 references
  1. [9]

    Z. Wei, S. Link, Towards the efficient discovery of meaningful functional depen- dencies, Information Systems 116 (2023) 102224. doi:10.1016/j.is.2023. 102224

    2023 doi

  2. [10]

    M. F. D. R., S. Groen, F. Panse, W. Wingerath, Navigating Tabular Data Synthesis Research: Understanding User Needs and Tool Capabilities (May 2024). doi: 10.48550/arXiv.2405.20959

    2024 doi arXiv

  3. [11]

    Goodfellow, J

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, Y . Bengio, Generative Adversarial Nets, in: Advances in Neu- ral Information Processing Systems, V ol. 27, Curran Associates, Inc., 2014, p. 139–144. doi:10.1145/3422622

    2014 doi

  4. [12]

    M. Sami, I. Mobin, A Comparative Study on Variational Autoencoders and Generative Adversarial Networks, in: 2019 International Conference of Artifi- cial Intelligence and Information Technology (ICAIIT), IEEE, 2019, pp. 1–5. doi:10.1109/ICAIIT.2019.8834544

    2019

  5. [13]

    Sattarov, M

    T. Sattarov, M. Schreyer, D. Borth, FinDiff: Diffusion Models for Financial Tab- ular Data Generation, in: 4th ACM International Conference on AI in Finance, ACM, 2023, pp. 64–72. doi:10.1145/3604237.3626876

    2023

  6. [14]

    Mahendra, C

    M. Mahendra, C. Umesh, S. Bej, K. Schultz, O. Wolkenhauer, Convex space learning for tabular synthetic data generation (Jul. 2024). doi:10.48550/ arXiv.2407.09789

    2024 doi arXiv

  7. [15]

    W. X. Zhao, K. Zhou, J. Li, T. Tang, X. Wang, Y . Hou, Y . Min, B. Zhang, J. Zhang, Z. Dong, Y . Du, C. Yang, Y . Chen, Z. Chen, J. Jiang, R. Ren, Y . Li, X. Tang, 20 Z. Liu, P. Liu, J.-Y . Nie, J.-R. Wen, A Survey of Large Language Models (Nov. 2023). doi:10.48550/arXiv.2303.18223

    2023 doi arXiv

  8. [16]

    Figueira, B

    A. Figueira, B. Vaz, Survey on Synthetic Data Generation, Evaluation Methods and GANs, Mathematics 10 (15) (2022) 2733. doi:10.3390/math10152733

    2022 doi

  9. [17]

    L. Xu, M. Skoularidou, A. Cuesta-Infante, K. Veeramachaneni, Modeling Tabular data using Conditional GAN, in: Advances in Neural Information Processing Systems, V ol. 32, Curran Associates, Inc., 2019

    2019

  10. [18]

    Schultz, S

    K. Schultz, S. Bej, W. Hahn, M. Wolfien, P. Srivastava, O. Wolkenhauer, Con- vGeN: A convex space learning approach for deep-generative oversampling and imbalanced classification of small tabular datasets, Pattern Recognition 147 (2024) 110138. doi:10.1016/j.patcog.2023.110138

    2024

  11. [19]

    Z. Lin, A. Khetan, G. Fanti, S. Oh, Pacgan: The power of two samples in genera- tive adversarial networks, IEEE Journal on Selected Areas in Information Theory 1 (1) (2020) 324–335. doi:10.1109/JSAIT.2020.2983071

    2020

  12. [20]

    Z. Zhao, A. Kunar, R. Birke, H. Van der Scheer, L. Y . Chen, CTAB-GAN +: enhancing tabular data synthesis, Frontiers in Big Data 6 (Jan. 2024). doi:10. 3389/fdata.2023.1296508

    2024

  13. [21]

    Weng, From GAN to WGAN (Apr

    L. Weng, From GAN to WGAN (Apr. 2019). doi:10.48550/arXiv.1904. 08994

    2019 doi

  14. [22]

    Vahdat, J

    A. Vahdat, J. Kautz, NV AE: A Deep Hierarchical Variational Autoencoder, in: Advances in Neural Information Processing Systems, V ol. 33, Curran Associates, Inc., 2020, pp. 19667–19679

    2020

  15. [23]

    D. P. Kingma, M. Welling, An Introduction to Variational Autoencoders, Foun- dations and Trends in Machine Learning 12 (4) (2019) 307–392. doi:10.1561/ 2200000056

    2019

  16. [24]

    S. Bej, C. Umesh, M. Mahendra, K. Schultz, J. Sarkar, O. Wolkenhauer, Ac- counting for diverse feature-types improves patient stratification on tabular clin- 21 ical datasets, Machine Learning with Applications 14 (2023) 100490. doi: 10.1016/j.mlwa.2023.100490

    2023

  17. [25]

    Borisov, K

    V . Borisov, K. Seßler, T. Leemann, M. Pawelczyk, G. Kasneci, Language Mod- els are Realistic Tabular Data Generators (Apr. 2023). doi:10.48550/arXiv. 2210.06280

    2023 doi arXiv

  18. [26]

    A. V . Solatorio, O. Dupriez, REaLTabFormer: Generating Realistic Relational and Tabular Data using Transformers (Feb. 2023). doi:10.48550/arXiv. 2302.02041

    2023 doi arXiv

  19. [27]

    Z. Zhao, R. Birke, L. Y . Chen, TabuLa: Harnessing Language Models for Tab- ular Data Synthesis, in: Advances in Knowledge Discovery and Data Mining, Springer Nature, 2025, pp. 247–259. doi:10.1007/978-981-96-8186-0_20

    2025 doi

  20. [28]

    T. Liu, Z. Qian, J. Berrevoets, M. v. d. Schaar, GOGGLE: Generative Modelling for Tabular Data by Learning Relational Structure, in: International Conference on Learning Representations, 2023

    2023

  21. [29]

    Zhang, J

    H. Zhang, J. Zhang, B. Srinivasan, Z. Shen, X. Qin, C. Faloutsos, H. Rangwala, G. Karypis, Mixed-Type Tabular Data Synthesis with Score-based Di ffusion in Latent Space, arXiv (2023). doi:10.48550/ARXIV.2310.09656

    2023 doi arXiv

  22. [30]

    J. Shi, M. Xu, H. Hua, H. Zhang, S. Ermon, J. Leskovec, TabDiff: a Multi-Modal Diffusion Model for Tabular Data Generation (Oct. 2024). doi:10.48550/ arXiv.2410.20626

    2024 doi arXiv

  23. [31]

    M. Vero, M. Balunovic, M. Vechev, CuTS: Customizable Tabular Synthetic Data Generation, in: International Conference on Machine Learning, 2024

    2024

  24. [32]

    C. Ge, S. Mohapatra, X. He, I. F. Ilyas, Kamino: Constraint-Aware Differentially Private Data Synthesis (Apr. 2021). doi:10.48550/arXiv.2012.15713

    2021 doi arXiv

  25. [33]

    P. Han, W. Xu, W. Lin, J. Cao, C. Liu, S. Duan, H. Zhu, C3-TGAN- Control- lable Tabular Data Synthesis with Explicit Correlations and Property Constraints, TechRxiv (Oct. 2023). doi:10.36227/techrxiv.24249643.v1. 22

    2023 doi

  26. [34]

    Zhang, Z

    J.-C. Zhang, Z. Zhou, Y .-J. Xiong, C.-M. Xia, F. Dai, CausalDiffTab: Mixed-Type Causal-Aware Diffusion for Tabular Data Generation, ArXiv abs /2506.14206 (Jun. 2025). doi:10.48550/arXiv.2506.14206. 23

    2025 doi arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.