Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

TabFSBench: Tabular Benchmark for Feature Shifts in Open Environments

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read TabFSBench introduces the first tabular feature-shift benchmark and reports a linear link between removed-feature importance and performance loss.

desk verdict Solid benchmark resource, but the headline linearity claim in Observation 2 is confounded by shift fraction and needs a controlled reanalysis. read the letter →

arxiv 2501.18935 v3 pith:PRTGX7VU submitted 2025-01-31 cs.LG

Zi-Jian Cheng , Zi-Yi Jia , Zhi Zhou , Yu-Feng Li , Lan-Zhe Guo This is my paper · ORCID
classification cs.LG
keywords featureshifttabulardataopenenvironmentsbenchmarkimportancemodelrobustnessPearsoncorrelationLLM
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 argues that feature shifts—changes in the set of input columns between training and deployment—are a distinct, under-studied failure mode for tabular machine learning. It introduces TabFSBench, a benchmark with twelve datasets and four shift protocols, and evaluates 24 models spanning tree ensembles, deep networks, a large language model, and tabular large language models. Its central empirical finding is that the summed Pearson correlation of the removed features tracks the model's relative performance drop, with $\rho = 0.74$ in the aggregated random-shift experiment. If the finding is right, feature importance becomes a practical predictor of deployment risk, and protecting strongly correlated features becomes a concrete robustness strategy.

What carries the argument

The load-bearing mechanism is a controlled removal protocol: rank features by $|\rho|$ (Pearson correlation with the target), remove them singly or in ascending/descending/random order, impute the missing columns with training-set mean (mode for categorical features), and measure the relative performance gap $\Delta = (\mathrm{metric}_i - \mathrm{metric}_0)/\mathrm{metric}_0$. The sum of the removed features' correlations serves as the importance index, and the benchmark API turns any dataset/model pair into a shift experiment. This machinery isolates feature removal as the only change between training and test, so that any observed degradation is attributable to the shift itself.

What would settle it

Run the comparison at a fixed shift fraction: on each dataset, remove the 20% highest-correlation features and separately remove the 20% lowest-correlation features, then compare the average performance gap. If the linear-importance claim is substantive, high-correlation removal must degrade performance clearly more than low-correlation removal at the same fraction; if the gap disappears once shift fraction is held fixed, Observation 2 is a monotonicity artifact.

Watch

Extended reading notes

Core claim

The paper's central claim is that TabFSBench is the first benchmark built specifically for feature shifts in tabular data, and that its experiments establish a quantitative relationship between the importance of a shifted feature set and model performance degradation. Feature importance is operationalized by the absolute Pearson correlation of each feature with the target, and shifted features are filled with training-set means before evaluation. Across single-feature, most-relevant, least-relevant, and random removal experiments, degradation grows as the correlation sum of the removed feature set grows; the paper reports $\rho = 0.74$ for this relationship. It also reports that most models lose accuracy steadily as shift degree increases, that no single model consistently dominates, and that closed-environment performance correlates with feature-shift performance.

Load-bearing premise

The random-shift experiment treats the correlation between the removed-feature correlation sum and the performance gap as evidence about feature importance, but both quantities increase automatically as more features are removed, so $\rho = 0.74$ may reflect shift size rather than a genuine linear importance–degradation law.

Editorial extensions

If this is right

  • Feature importance rankings derived from Pearson correlation can be used to estimate expected deployment loss for a given feature-removal budget.
  • Because most tree-based and deep models degrade under feature decrement, robust tabular systems should handle missing columns explicitly rather than relying on fixed architectures.
  • Tabular large language models show relatively small performance gaps on classification tasks, pointing to prompt- or fine-tuning-based defenses as a promising direction.
  • Closed-environment accuracy correlates with feature-shift robustness, so improving baseline performance is unlikely to harm shift resilience and may help it.
  • The benchmark's four shift protocols, twelve datasets, and 24 models provide a reusable evaluation standard for future feature-shift methods.

Reading between the lines

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

  • The reported $\rho = 0.74$ is likely inflated by the experimental design, since both the correlation sum and the performance gap are monotone in the number of removed features; controlling for shift fraction within datasets would give a cleaner test of the importance–degradation link.
  • If the importance–degradation relationship is causal, a direct implication is that removing a few high-correlation features should hurt as much as removing many low-correlation features at the same shift degree—an experiment the paper does not run.
  • The choice of mean imputation is itself a confound; re-running the benchmark with learned imputation or generative feature reconstruction could strengthen or weaken the observed linearity.
  • Extending the protocol to feature increments, rather than decrements only, would test whether the same importance law governs the opposite direction of shift.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper introduces TabFSBench, a benchmark for tabular feature shifts, comprising twelve datasets from binary/multi-class classification and regression, four feature-shift scenarios (single, most/least-relevant, random), and an evaluation of four model categories including LLMs and tabular LLMs. It reports three observations: most models show limited robustness under feature shifts; shifted-feature importance has a claimed linear relationship with performance degradation (Section 5, Figure 4, rho=0.74); and closed-environment performance correlates with feature-shift performance. The paper provides code, APIs, and a leaderboard.

Significance. If the empirical claims are supported, this is a useful public resource for a relatively under-studied problem: the benchmark design, the breadth of models evaluated, and the release of callable APIs are concrete strengths, and the paper is the first systematic tabular feature-shift benchmark as far as the related-work discussion shows. The most interesting quantitative claim, Observation 2, is also the most fragile, and the current evidence for a linear importance-degradation relationship is not yet convincing. The benchmark itself remains valuable even if Observation 2 is weakened, and the paper's three-seed averaging and per-dataset appendices are commendable for reproducibility.

major comments (4)
  1. [§5, Observation 2 (Figure 4)] The claim 'shifted feature set importance has a linear relationship with model performance degradation' is confounded by shift fraction. In the random-shift protocol of Section 3.2, the sum of absolute Pearson correlations of removed features and the performance gap Δ both increase monotonically with the number of features removed. The scatter in Figure 4 pools all shift ratios and all datasets, so the Pearson correlation rho=0.74 can emerge even if, at any fixed shift ratio, the identities of the removed features do not matter at all. The paper does not report a partial correlation controlling for shift ratio, nor a within-ratio analysis (e.g., comparing most-relevant versus least-relevant removal at the same degree). The adjacent most/least-relevant experiment (Figure 9) partially addresses this, but Figure 4's pooled analysis does not; please re-analyze within fixed shift ratios or report partial correlations, and temper the 'linear' wording accordingly.
  2. [§5, Observation 2 (Figure 4)] A Pearson correlation of 0.74 between two monotone quantities does not establish linearity, especially when the points are not independent samples but rather pooled trajectories from twelve datasets. The manuscript calls this a 'linear trend' and uses it to motivate future feature-importance-driven optimization. Please provide per-dataset slopes and confidence intervals, and test whether residuals after removing the shift-ratio effect are still informative. If the relationship is only ordinal, the conclusion should be stated as ordinal monotonicity rather than linearity.
  3. [§3.2, Table 2 and Appendix G.1] The statement that 'the marginal differences between these metrics' rankings are found to be statistically insignificant' is not supported by any test or confidence interval in the text. Since the choice of Pearson correlation as the sole importance metric affects the main analysis, please give the sample size, the test used, and the resulting p-value or interval, or remove the significance claim.
  4. [§4.2 and Appendix H.1] All results are reported as averages over three random seeds, but no error bars, standard deviations, or per-seed results are shown in the main figures. Several observations (e.g., 'no model consistently outperforms' in Table 4, and the robustness ordering in Table 3) depend on small differences in averaged values; without variance information, some of these distinctions may not be meaningful. Adding error bars or a statistical comparison for the headline rankings would strengthen the benchmark conclusions.
minor comments (5)
  1. [§4.1 and Appendix E.1] Reference errors: XGBoost is cited as (Chizat et al., 2020) and LightGBM as (Badirli et al., 2020), but Chizat et al. is the Sinkhorn divergence paper and Badirli et al. is GrowNet; the correct citations are Chen & Guestrin (2016) and Ke et al. (2017). DCNv2 is described via the Deformable ConvNets reference (Zhu et al., 2019), while the tabular DCNv2 used in the experiments is Wang et al. (2021b).
  2. [§3.2, Single Shift] The description 'we sequentially remove one feature by employing a sampling-with-replacement approach in the ascending order' is unclear: sampling with replacement and 'ascending order' are incompatible. Please clarify whether features are removed independently in correlation order or sampled with replacement for each trial.
  3. [Figure 4 caption] The caption says 'Sum of shifted feature set's correlations' but does not state whether absolute values are used; this matters because correlations can be negative. Please specify the exact quantity used (e.g., sum of |PCC|) in the caption and in Section 3.2.
  4. [Appendix H.7] The appendix headers use inconsistent acronyms (SC, MC-D, MC-A, MC-R in the text, but MC-M, MC-L, RD in the tables). Please unify the terminology so that the random-shift tables are labeled consistently with the main text.
  5. [Section 6, Limitations] The paper's own limitation list is useful, but the third limitation ('does not examine how models respond to shifted features with identical correlation structures') directly interacts with Observation 2, since identical-correlation features are exactly the cases that could separate importance from shift-ratio effects. Please connect this limitation to the analysis in Figure 4.

Circularity Check

1 steps flagged · score 6.0 of 10

Observation 2's 'linear relationship' between shifted-feature correlation sum and performance degradation is built into the random-shift construction: both quantities are monotone functions of shift ratio, so rho = 0.74 largely measures shift magnitude rather than feature importance.

  1. other [Section 3.2 (Random Shift protocol); Section 5, Observation 2; Figure 4 caption]
    "By progressively removing features during testing (where removing one feature represents a shift ratio of 1/n, with n being the total feature count), we quantify performance degradation across different shift magnitudes. ... We observe a linear trend between the correlations of the shifted feature set and model performance degradation (see the blue line in Figure 4; ρ = 0.74)."

    In the random-shift protocol, both the x-axis and y-axis of Figure 4 are generated by increasing the shift ratio. Removing more features mechanically increases the sum of absolute Pearson correlations of the removed set, and Table 3 shows that the performance gap also grows monotonically with shift degree. The reported Pearson correlation pools across all shift ratios and datasets, so the positive association is largely a consequence of the common manipulated variable, shift magnitude, rather than an independent property of feature importance.

full rationale

TabFSBench itself is a self-contained benchmark contribution: it defines four feature-shift protocols, evaluates many models, and provides an API, and its benchmark infrastructure does not depend on a circular derivation. I found no load-bearing self-citations: references to the authors' earlier work are contextual, not used as a uniqueness theorem or as justification for a fitted choice. The circularity concern is concentrated in the paper's central analytical observation, Observation 2. Figure 4 computes Pearson correlation between the sum of correlations of shifted features and Delta, using points that vary simultaneously in the number of shifted features. Since both quantities are monotone increasing functions of shift ratio, the strong correlation rho = 0.7405 is substantially predetermined by the random-shift construction; it does not establish that, at a fixed shift degree, more important features cause more degradation. The separate most/least-relevant experiment (Figure 9) does compare importance at a fixed degree and supports a directional effect, which is why the circularity is partial rather than total. Observations 1 and 3 are empirical summaries with independent content. Overall, the benchmark is not circular, but the headline quantitative claim of a linear importance-degradation relationship is partially circular because it reduces, by construction, to the shift-magnitude axis of the experimental design.

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

The central claims rest on domain assumptions about how to simulate feature shifts and how to measure feature importance. No new physical entities or fitted free parameters are introduced. The main concern is the unstated confounding in Observation 2, which is captured in the red flags.

assumptions (4)
  • domain assumption Feature importance can be measured by the absolute Pearson correlation between each feature and the target.
    Section 3.2 chooses PCC over Spearman, SHAP, and mutual information; the entire importance-degradation analysis depends on this choice.
  • domain assumption Mean imputation is a neutral way to simulate feature shifts and does not introduce artificial distribution shift.
    Section 3.3 selects mean-value imputation; the performance gaps are interpreted as robustness to feature shift rather than artifacts of imputation.
  • domain assumption Feature increment scenarios are unimportant because models truncate new features and performance does not degrade.
    Section 2.2 justifies focusing only on feature decrement based on this claim, which is asserted without experiments.
  • domain assumption The 12 selected datasets are representative of tabular tasks in open environments.
    Section 3.1 chooses datasets from OpenML and Kaggle; the generality of all three observations rests on this representativeness.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TabFSBench: Tabular Benchmark for Feature Shifts in Open Environments." pith.science (2026). https://pith.science/paper/PRTGX7VU

@misc{pith2026250118935,
  author       = {Pith},
  title        = {Pith review of: TabFSBench: Tabular Benchmark for Feature Shifts in Open Environments},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PRTGX7VU}},
  note         = {Machine review of arXiv:2501.18935}
}
read the original abstract

Tabular data is widely utilized in various machine learning tasks. Current tabular learning research predominantly focuses on closed environments, while in real-world applications, open environments are often encountered, where distribution and feature shifts occur, leading to significant degradation in model performance. Previous research has primarily concentrated on mitigating distribution shifts, whereas feature shifts, a distinctive and unexplored challenge of tabular data, have garnered limited attention. To this end, this paper conducts the first comprehensive study on feature shifts in tabular data and introduces the first tabular feature-shift benchmark (TabFSBench). TabFSBench evaluates impacts of four distinct feature-shift scenarios on four tabular model categories across various datasets and assesses the performance of large language models (LLMs) and tabular LLMs in the tabular benchmark for the first time. Our study demonstrates three main observations: (1) most tabular models have the limited applicability in feature-shift scenarios; (2) the shifted feature set importance has a linear relationship with model performance degradation; (3) model performance in closed environments correlates with feature-shift performance. Future research direction is also explored for each observation. Benchmark: https://github.com/LAMDASZ-ML/TabFSBench.

Figures

Figures reproduced from arXiv: 2501.18935 by the authors.

Figure 1
Figure 1. Open environment challenges: a) Distribution Shifts. Change in data distribution between training and testing while keeping features unchanged. b) Feature Shifts. Change in the feature set between training and testing, either by addition or removal of features. generalization (Zhou et al., 2022; Zhao et al., 2024). Vari￾ous benchmarks on distribution shifts in tabular data have been established. For example, TableSh… view at source ↗
Figure 2
Figure 2. Upper bound and original model performance. to be shifted. The valid range for this parameter is from 0 to 1, where 0 signifies no features are removed, and 1 signifies all features are removed. The export_dataset parameter controls whether the modified dataset (after removing shifted features) is ex￾ported as a CSV file for further use. An example command for running a feature-shift experiment in TabFSBench is as f… view at source ↗
Figure 3
Figure 3. Results for single shift experiments with correlation analysis experiments across 12 datasets. The x-axis represents features ordered in ascending correlation with the target, the left y-axis represents the average performance (accuracy or RMSE) of all models and the right y-axis shows the absolute value of correlation. Features are removed in ascending order of their correlation values to observe the impact on mode… view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: We use ∆ (described in equation 1) to measure the per￾formance decrease. Sum of shifted feature set’s correlations refers to the sum of Pearson correlation coefficients of shifted features. Notably, performance decrease and sum of shifted feature set’s correlations dem…
Figure 6
Figure 6. Figure 6: Pearson Correlation Analysis of the Datasets. Images are sorted in binary classification, multi-class classification, and regression order. The last column of each image is chosen as the input column with the target Pearson correlation coefficient. 18 [PITH_FULL_IMAGE…
Figure 7
Figure 7. Figure 7: LLM Prompt for the experiments. Take the Penguins dataset as an example. across a wide range of datasets. F. Feature Imputation Methods We compared the performance of various models using their own imputation methods, random imputation, and mean imputation [PITH_FULL_…
Figure 8
Figure 8. Figure 8: Visual comparison of feature importance rankings across Pearson correlation, Spearman correlation, mutual information, and SHAP values on twelve datasets. 24 [PITH_FULL_IMAGE:figures/full_fig_p024_8.png]
Figure 9
Figure 9. Figure 9: Results for most/least shift experiments with correlation analysis across datasets. The x-axis represents the percentage of features removed, while the y-axis indicates the average performance (accuracy or RMSE) of all models. Three feature removal strategies are compa…
Figure 10
Figure 10. Figure 10: Trajectories between correlation and accuracy for each of twelve datasets. 30 [PITH_FULL_IMAGE:figures/full_fig_p030_10.png]
Figure 11
Figure 11. Figure 11: Average feature-shift performance vs. average runtime for each model. 31 [PITH_FULL_IMAGE:figures/full_fig_p031_11.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Realistic Evaluation of TabPFN v2 in Open Environments

    cs.LG 2025-05 conditional novelty 5.0 of 10

    TabPFN v2 underperforms tree-based models on most open-environment tabular tasks and is only preferable on small, covariate-shifted, class-balanced data.

Reference graph

Works this paper leans on

72 extracted references · 51 canonical work pages · cited by 1 Pith paper

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    Domain-adversarial neural networks

    Ajakan, H., Germain, P., Larochelle, H., Laviolette, F., and Marchand, M. Domain-adversarial neural networks. arXiv preprint arXiv:1412.4446, 2014

    arXiv 2014

  3. [3]

    and Krzywinski, M

    Altman, N. and Krzywinski, M. Tabular data. Nature Methods, 14 0 (4): 0 329--331, 2017

    work page 2017

  4. [4]

    Arik, S. \"O . and Pfister, T. Tabnet: Attentive interpretable tabular learning . In Proceedings of the 35th AAAI Conference on Artificial Intelligence, pp.\ 6679--6687, 2021

    work page 2021

  5. [5]

    Invariant risk minimization

    Arjovsky, M., Bottou, L., Gulrajani, I., and Lopez-Paz, D. Invariant risk minimization. arXiv preprint arXiv:1907.02893, 2019

    arXiv 1907

  6. [6]

    Badirli, S., Liu, X., Xing, Z., Bhowmik, A., Doan, K., and Keerthi, S. K. Gradient boosting neural networks: Grownet . arXiv preprint arXiv:2002.07971, 2020

    arXiv 2002

  7. [7]

    It's COMPASlicated: The Messy Relationship between RAI Datasets and Algorithmic Fairness Benchmarks

    Bao, M., Zhou, A., Zottola, S., Brubach, B., Desmarais, S., Horowitz, A., Lum, K., and Venkatasubramanian, S. It's COMPASlicated: The Messy Relationship between RAI Datasets and Algorithmic Fairness Benchmarks . arXiv preprint arXiv:2106.05498, 2021

    arXiv 2021

  8. [8]

    Deep neural networks and tabular data: A survey

    Borisov, V., Leemann, T., Se ler, K., Haug, J., Pawelczyk, M., and Kasneci, G. Deep neural networks and tabular data: A survey . IEEE Transactions on Neural Networks and Learning Systems, 35 0 (6): 0 7499--7519, 2022

    work page 2022

Show all 72 references
  1. [9]

    Z., and Wu, J

    Chen, J., Liao, K., Wan, Y., Chen, D. Z., and Wu, J. Danets: Deep abstract networks for tabular data classification and regression . In Proceedings of the 36th AAAI Conference on Artificial Intelligence, pp.\ 3930--3938, 2022

    2022

  2. [10]

    Tabcaps: A capsule neural network for tabular data classification with bow routing

    Chen, J., Liao, K., Fang, Y., Chen, D., and Wu, J. Tabcaps: A capsule neural network for tabular data classification with bow routing . In Proceedings of the 11th International Conference on Learning Representations, 2023

    2023

  3. [11]

    and Guestrin, C

    Chen, T. and Guestrin, C. XGBoost: A Scalable Tree Boosting System . In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp.\ 785--794, 2016

    2016

  4. [12]

    Faster Wasserstein distance estimation with the Sinkhorn divergence

    Chizat, L., Roussillon, P., L \'e ger, F., Vialard, F.-X., and Peyr \'e , G. Faster Wasserstein distance estimation with the Sinkhorn divergence . Advances in Neural Information Processing Systems, pp.\ 2257--2269, 2020

    2020

  5. [13]

    ImageNet: A large-scale hierarchical image database

    Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., and Fei-Fei, L. ImageNet: A large-scale hierarchical image database . In Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition, pp.\ 248--255, 2009

    2009

  6. [14]

    A., Zhang, J., Hu, Z., Qi, Y

    Fang, X., Xu, W., Tan, F. A., Zhang, J., Hu, Z., Qi, Y. J., Nickleach, S., Socolinsky, D., Sengamedu, S., and Faloutsos, C. Large language models on tabular data: Prediction, generation, and understanding-a survey . arXiv preprint arXiv:2402.17944, 2024

    2024 arXiv

  7. [15]

    C., and Schmidt, L

    Gardner, J., Perdomo, J. C., and Schmidt, L. Large Scale Transfer Learning for Tabular Data via Language Modeling . arXiv preprint arXiv:2406.12031, 2024 a

    2024 arXiv

  8. [16]

    Benchmarking distribution shift in tabular data with tableshift

    Gardner, J., Popovic, Z., and Schmidt, L. Benchmarking distribution shift in tabular data with tableshift . Advances in Neural Information Processing Systems, pp.\ 53385--53432, 2024 b

    2024

  9. [17]

    F., Ellis, D

    Gemmeke, J. F., Ellis, D. P., Freedman, D., Jansen, A., Lawrence, W., Moore, R. C., Plakal, M., and Ritter, M. Audio set: An ontology and human-labeled dataset for audio events . In Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, pp...

    2017

  10. [18]

    Revisiting deep learning models for tabular data

    Gorishniy, Y., Rubachev, I., Khrulkov, V., and Babenko, A. Revisiting deep learning models for tabular data . Advances in Neural Information Processing Systems, pp.\ 18932--18943, 2021

    2021

  11. [19]

    TabR: Tabular Deep Learning Meets Nearest Neighbors

    Gorishniy, Y., Rubachev, I., Kartashev, N., Shlenskii, D., Kotelnikov, A., and Babenko, A. TabR: Tabular Deep Learning Meets Nearest Neighbors . In Proceedings of the 12th International Conference on Learning Representations, 2024

    2024

  12. [20]

    Why do tree-based models still outperform deep learning on typical tabular data? Advances in Neural Information Processing Systems, pp.\ 507--520, 2022

    Grinsztajn, L., Oyallon, E., and Varoquaux, G. Why do tree-based models still outperform deep learning on typical tabular data? Advances in Neural Information Processing Systems, pp.\ 507--520, 2022

    2022

  13. [21]

    Robust semi-supervised learning in open environments

    Guo, L.-Z., Jia, L.-H., Shao, J.-J., and Li, Y.-F. Robust semi-supervised learning in open environments. Frontiers of Computer Science, 19 0 (8): 0 198345, 2025

    2025

  14. [22]

    Domain adaptation for time series under feature and label shifts

    He, H., Queen, O., Koker, T., Cuevas, C., Tsiligkaridis, T., and Zitnik, M. Domain adaptation for time series under feature and label shifts. In Proceedings of the 40th International Conference on Machine Learning, pp.\ 12746--12774, 2023

    2023

  15. [23]

    TabLLM: Few-shot Classification of Tabular Data with Large Language Models

    Hegselmann, S., Buendia, A., Lang, H., Agrawal, M., Jiang, X., and Sontag, D. TabLLM: Few-shot Classification of Tabular Data with Large Language Models . In Proceedings of the 26th International Conference on Artificial Intelligence and Statistics, pp.\ 5549--5581, 2023

    2023

  16. [24]

    Tab PFN : A Transformer That Solves Small Tabular Classification Problems in a Second

    Hollmann, N., M \"u ller, S., Eggensperger, K., and Hutter, F. Tab PFN : A Transformer That Solves Small Tabular Classification Problems in a Second . In Proceedings of the 11th International Conference on Learning Representations, 2023 a

    2023

  17. [25]

    Large language models for automated data science: Introducing caafe for context-aware automated feature engineering

    Hollmann, N., M \"u ller, S., and Hutter, F. Large language models for automated data science: Introducing caafe for context-aware automated feature engineering . Advances in Neural Information Processing Systems, pp.\ 44753--44775, 2023 b

    2023

  18. [26]

    Tabtransformer: Tabular data modeling using contextual embeddings

    Huang, X., Khetan, A., Cvitkovic, M., and Karnin, Z. Tabtransformer: Tabular data modeling using contextual embeddings . arXiv preprint arXiv:2012.06678, 2020

    2012 arXiv

  19. [27]

    Iversen, G. R. and Gergen, M. Statistics: The conceptual approach . Springer Science & Business Media, 2012

    2012

  20. [28]

    TANGOS : Regularizing Tabular Neural Networks through Gradient Orthogonalization and Specialization

    Jeffares, A., Liu, T., Crabb \'e , J., Imrie, F., and van der Schaar, M. TANGOS : Regularizing Tabular Neural Networks through Gradient Orthogonalization and Specialization . In Proceedings of the 11th International Conference on Learning Representations, 2023

    2023

  21. [29]

    LAMDA-SSL: a comprehensive semi-supervised learning toolkit

    Jia, L.-H., Guo, L.-Z., Zhou, Z., and Li, Y.-F. LAMDA-SSL: a comprehensive semi-supervised learning toolkit . Science China. Information Sciences, 67 0 (1): 0 117101, 2024 a

    2024

  22. [30]

    Realistic evaluation of semi-supervised learning algorithms in open environments

    Jia, L.-H., Guo, L.-Z., Zhou, Z., and Li, Y.-F. Realistic evaluation of semi-supervised learning algorithms in open environments. In Proceedings of the 12th International Conference on Learning Representations, 2024 b

    2024

  23. [31]

    Well-tuned simple nets excel on tabular datasets

    Kadra, A., Lindauer, M., Hutter, F., and Grabocka, J. Well-tuned simple nets excel on tabular datasets . Advances in Neural Information Processing Systems, pp.\ 23928--23941, 2021

    2021

  24. [32]

    LightGBM: A Highly Efficient Gradient Boosting Decision Tree

    Ke, G., Meng, Q., Finley, T., Wang, T., Chen, W., Ma, W., Ye, Q., and Liu, T.-Y. LightGBM: A Highly Efficient Gradient Boosting Decision Tree . Advances in Neural Information Processing Systems, pp.\ 3149--3157, 2017

    2017

  25. [33]

    Self-normalizing neural networks

    Klambauer, G., Unterthiner, T., Mayr, A., and Hochreiter, S. Self-normalizing neural networks . Advances in Neural Information Processing Systems, pp.\ 972–--981, 2017

    2017

  26. [34]

    Wild-Tab: A Benchmark For Out-Of-Distribution Generalization In Tabular Regression

    Kolesnikov, S. Wild-Tab: A Benchmark For Out-Of-Distribution Generalization In Tabular Regression . arXiv preprint arXiv:2312.01792, 2023

    2023 arXiv

  27. [35]

    A comprehensive survey on test-time adaptation under distribution shifts

    Liang, J., He, R., and Tan, T. A comprehensive survey on test-time adaptation under distribution shifts. International Journal of Computer Vision, pp.\ 1--34, 2024

    2024

  28. [36]

    On the need for a language describing distribution shifts: Illustrations on tabular datasets

    Liu, J., Wang, T., Cui, P., and Namkoong, H. On the need for a language describing distribution shifts: Illustrations on tabular datasets . Advances in Neural Information Processing Systems, pp.\ 51371--51408, 2023

    2023

  29. [37]

    TALENT: A Tabular Analytics and Learning Toolbox

    Liu, S.-Y., Cai, H.-R., Zhou, Q.-L., and Ye, H.-J. TALENT: A Tabular Analytics and Learning Toolbox . arXiv preprint arXiv:2407.04057, 2024

    2024 arXiv

  30. [38]

    Shifts: A Dataset of Real Distributional Shift Across Multiple Large-Scale Tasks

    Malinin, A., Band, N., Gal, Y., Gales, M., Ganshin, A., Chesnokov, G., Noskov, A., Ploskonosov, A., Prokhorenkova, L., Provilkov, I., Raina, V., Raina, V., Roginskiy, D., Shmatova, M., Tigas, P., and Yangel, B. Shifts: A Dataset of Real Distributional Shift Across Multiple Lar...

    2021

  31. [39]

    B., Gales, M

    Malinin, A., Athanasopoulos, A., Barakovic, M., Cuadra, M. B., Gales, M. J. F., Granziera, C., Graziani, M., Kartashev, N., Kyriakopoulos, K., Lu, P.-J., Molchanova, N., Nikitakis, A., Raina, V., Rosa, F. L., Sivena, E., Tsarsitalidis, V., Tsompopoulou, E., and Volf, E. Shifts...

    2022 arXiv

  32. [40]

    When do neural nets outperform boosted trees on tabular data? Advances in Neural Information Processing Systems, pp.\ 76336--76369, 2023

    McElfresh, D., Khandagale, S., Valverde, J., Prasad C, V., Ramakrishnan, G., Goldblum, M., and White, C. When do neural nets outperform boosted trees on tabular data? Advances in Neural Information Processing Systems, pp.\ 76336--76369, 2023

    2023

  33. [41]

    Uncertainty estimation for classification and risk prediction on medical tabular data

    Meijerink, L., Cin \`a , G., and Tonutti, M. Uncertainty estimation for classification and risk prediction on medical tabular data. arXiv preprint arXiv:2004.05824, 2020

    2004 arXiv

  34. [42]

    The effect of natural distribution shift on question answering models

    Miller, J., Krauth, K., Recht, B., and Schmidt, L. The effect of natural distribution shift on question answering models . In Proceedings of the 37th International Conference on Machine Learning, pp.\ 6905--6916, 2020

    2020

  35. [43]

    Optimized Feature Generation for Tabular Data via LLMs with Decision Tree Reasoning

    Nam, J., Kim, K., Oh, S., Tack, J., Kim, J., and Shin, J. Optimized Feature Generation for Tabular Data via LLMs with Decision Tree Reasoning . Advances in Neural Information Processing Systems, pp.\ 92352--92380, 2024

    2024

  36. [44]

    GPT-4 Technical Report

    OpenAI. GPT-4 Technical Report . arXiv preprint arXiv:2303.08774, 2024

    2024 arXiv

  37. [45]

    Open-world machine learning: applications, challenges, and opportunities

    Parmar, J., Chouhan, S., Raychoudhury, V., and Rathore, S. Open-world machine learning: applications, challenges, and opportunities. ACM Computing Surveys, 55 0 (10): 0 1--37, 2023

    2023

  38. [46]

    Neural Oblivious Decision Ensembles for Deep Learning on Tabular Data

    Popov, S., Morozov, S., and Babenko, A. Neural Oblivious Decision Ensembles for Deep Learning on Tabular Data . In Proceedings of the 8th International Conference on Learning Representations, 2020

    2020

  39. [47]

    V., and Gulin, A

    Prokhorenkova, L., Gusev, G., Vorobev, A., Dorogush, A. V., and Gulin, A. CatBoost: unbiased boosting with categorical features . Advances in Neural Information Processing Systems, pp.\ 6639–--6649, 2018

    2018

  40. [48]

    An overview of gradient descent optimization algorithms

    Ruder, S. An overview of gradient descent optimization algorithms . arXiv preprint arXiv:1609.04747, 2016

    2016 arXiv

  41. [49]

    Explainable artificial intelligence for tabular data: A survey

    Sahakyan, M., Aung, Z., and Rahwan, T. Explainable artificial intelligence for tabular data: A survey. IEEE access, 9: 0 135392--135422, 2021

    2021

  42. [50]

    Open-set learning under covariate shift

    Shao, J.-J., Yang, X.-W., and Guo, L.-Z. Open-set learning under covariate shift. Machine Learning, 113 0 (4): 0 1643--1659, 2024

    2024

  43. [51]

    Shen, M., Bu, Y., and Wornell, G. W. On balancing bias and variance in unsupervised multi-source-free domain adaptation . In Proceedings of the 40th International Conference on Machine Learning, pp.\ 30976--30991, 2023

    2023

  44. [52]

    B., and Goldstein, T

    Somepalli, G., Goldblum, M., Schwarzschild, A., Bruss, C. B., and Goldstein, T. Saint: Improved neural networks for tabular data via row attention and contrastive pre-training . arXiv preprint arXiv:2106.01342, 2021

    2021 arXiv

  45. [53]

    Autoint: Automatic feature interaction learning via self-attentive neural networks

    Song, W., Shi, C., Xiao, Z., Duan, Z., Xu, Y., Zhang, M., and Tang, J. Autoint: Automatic feature interaction learning via self-attentive neural networks . In Proceedings of the 28th ACM International Conference on Information and Knowledge Management, pp.\ 1161--1170, 2019

    2019

  46. [54]

    Llama: Open and efficient foundation language models

    Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M.-A., Lacroix, T., Rozi \`e re, B., Goyal, N., Hambro, E., Azhar, F., et al. Llama: Open and efficient foundation language models . arXiv preprint arXiv:2302.13971, 2023

    2023 arXiv

  47. [55]

    GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding

    Wang, A., Singh, A., Michael, J., Hill, F., Levy, O., and Bowman, S. GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding . In Proceedings of the 2018 EMNLP Workshop B lackbox NLP : Analyzing and Interpreting Neural Networks for NLP , pp.\ 353–...

    2018

  48. [56]

    Tent: Fully Test-Time Adaptation by Entropy Minimization

    Wang, D., Shelhamer, E., Liu, S., Olshausen, B., and Darrell, T. Tent: Fully Test-Time Adaptation by Entropy Minimization . In Proceedings of the 9th International Conference on Learning Representations, 2021 a

    2021

  49. [57]

    Dcn v2: Improved deep & cross network and practical lessons for web-scale learning to rank systems

    Wang, R., Shivanna, R., Cheng, D., Jain, S., Lin, D., Hong, L., and Chi, E. Dcn v2: Improved deep & cross network and practical lessons for web-scale learning to rank systems . In Proceedings of the Web Conference 2021, pp.\ 1785--1797, 2021 b

    2021

  50. [58]

    Unipredict: Large language models are universal tabular predictors

    Wang, R., Wang, Z., and Sun, J. Unipredict: Large language models are universal tabular predictors . arXiv preprint arXiv:2310.03266, 2023

    2023 arXiv

  51. [59]

    Usb: A unified semi-supervised learning benchmark for classification

    Wang, Y., Chen, H., Fan, Y., Sun, W., Tao, R., Hou, W., Wang, R., Yang, L., Zhou, Z., Guo, L.-Z., et al. Usb: A unified semi-supervised learning benchmark for classification . Advances in Neural Information Processing Systems, pp.\ 3938--3961, 2022

    2022

  52. [60]

    Neural network credit scoring models

    West, D. Neural network credit scoring models. Computers & operations research, 27 0 (11-12): 0 1131--1152, 2000

    2000

  53. [61]

    Switchtab: Switched autoencoders are effective tabular learners

    Wu, J., Chen, S., Zhao, Q., Sergazinov, R., Li, C., Liu, S., Zhao, C., Xie, T., Guo, H., and Ji, C. Switchtab: Switched autoencoders are effective tabular learners . In Proceedings of the 38th AAAI Conference on Artificial Intelligence, pp.\ 15924--15933, 2024

    2024

  54. [62]

    Modern Neighborhood Components Analysis: A Deep Tabular Baseline Two Decades Later

    Ye, H.-J., Yin, H.-H., and Zhan, D.-C. Modern Neighborhood Components Analysis: A Deep Tabular Baseline Two Decades Later . arXiv preprint arXiv:2407.03257, 2024

    2024 arXiv

  55. [63]

    Y ld z, A. Y. and Kalayci, A. Gradient Boosting Decision Trees on Medical Diagnosis over Tabular Data . arXiv preprint arXiv:2410.03705, 2024

    2024 arXiv

  56. [64]

    A., Fan, Z., Luo, H., Liu, F., Liu, Q., Cao, W., and Jian, L

    Zhang, T., Zhang, Z. A., Fan, Z., Luo, H., Liu, F., Liu, Q., Cao, W., and Jian, L. OpenFE: automated feature generation with expert-level performance . In Proceedings of the 40th International Conference on Machine Learning, pp.\ 41880--41901, 2023

    2023

  57. [65]

    Domain generalization for cross-domain fault diagnosis: An application-oriented perspective and a benchmark study

    Zhao, C., Zio, E., and Shen, W. Domain generalization for cross-domain fault diagnosis: An application-oriented perspective and a benchmark study . Reliability Engineering & System Safety, pp.\ 109964, 2024

    2024

  58. [66]

    Domain Generalization with MixStyle

    Zhou, K., Yang, Y., Qiao, Y., and Xiang, T. Domain Generalization with MixStyle . In Proceedings of the 9th International Conference on Learning Representations, 2021

    2021

  59. [67]

    Zhou, K., Liu, Z., Qiao, Y., Xiang, T., and Loy, C. C. Domain generalization: A survey . IEEE Transactions on Pattern Analysis and Machine Intelligence, 45 0 (4): 0 4396--4415, 2022

    2022

  60. [68]

    Ods: Test-time adaptation in the presence of open-world data shift

    Zhou, Z., Guo, L.-Z., Jia, L.-H., Zhang, D., and Li, Y.-F. Ods: Test-time adaptation in the presence of open-world data shift . In Proceedings of the 40th International Conference on Machine Learning, pp.\ 42574--42588, 2023

    2023

  61. [69]

    Fully Test-time Adaptation for Tabular Data

    Zhou, Z., Yang, Y.-K., Guo, L.-Z., and Li, Y.-F. Fully Test-time Adaptation for Tabular Data . In Proceedings of the 39th AAAI conference on Artificial Intelligence, 2025

    2025

  62. [70]

    Open-environment machine learning

    Zhou, Z.-H. Open-environment machine learning. National Science Review, 9 0 (8): 0 nwac123, 2022

    2022

  63. [71]

    TAT-QA: A question answering benchmark on a hybrid of tabular and textual content in finance

    Zhu, F., Lei, W., Huang, Y., Wang, C., Zhang, S., Lv, J., Feng, F., and Chua, T.-S. TAT-QA: A question answering benchmark on a hybrid of tabular and textual content in finance . arXiv preprint arXiv:2105.07624, 2021

    2021 arXiv

  64. [72]

    Deformable convnets v2: More deformable, better results

    Zhu, X., Hu, H., Lin, S., and Dai, J. Deformable convnets v2: More deformable, better results . In Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition, pp.\ 9308--9316, 2019

    2019

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.