uLEAD-TabPFN: Uncertainty-aware Dependency-based Anomaly Detection with TabPFN

Craig Xie; Jiuyong Li; Jixue Liu; Lin Liu; Sha Lu; Stefan Peters; Thuc Duy Le

arxiv: 2604.20255 · v1 · submitted 2026-04-22 · 💻 cs.LG · cs.AI

uLEAD-TabPFN: Uncertainty-aware Dependency-based Anomaly Detection with TabPFN

Sha Lu , Jixue Liu , Stefan Peters , Thuc Duy Le , Craig Xie , Lin Liu , Jiuyong Li This is my paper

Pith reviewed 2026-05-10 01:02 UTC · model grok-4.3

classification 💻 cs.LG cs.AI

keywords anomaly detectiontabular dataTabPFNdependency-based anomaly detectionuncertainty-awarehigh-dimensional dataADBenchPrior-Data Fitted Networks

0 comments

The pith

uLEAD-TabPFN detects tabular anomalies as violations of conditional dependencies modeled by frozen PFNs with uncertainty-aware scoring.

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

The paper introduces uLEAD-TabPFN, a framework for anomaly detection in tabular data that focuses on complex feature dependencies rather than proximity. It uses pre-trained Prior-Data Fitted Networks to estimate these dependencies in a latent space and scores anomalies based on uncertainty to handle noise. This leads to better performance on high-dimensional datasets compared to existing methods, as shown on 57 datasets from ADBench where it ranks highest on average. The approach addresses limitations of traditional methods that struggle with high dimensions and intricate dependencies.

Core claim

uLEAD-TabPFN identifies anomalies as violations of conditional dependencies in a learned latent space by leveraging frozen PFNs for dependency estimation combined with uncertainty-aware scoring, enabling robust and scalable detection particularly in medium- and high-dimensional tabular data.

What carries the argument

The uLEAD-TabPFN framework, which models conditional dependencies using frozen Prior-Data Fitted Networks (PFNs) in a latent space and applies uncertainty-aware scoring to flag anomalies.

If this is right

On high-dimensional datasets, uLEAD-TabPFN improves average ROC-AUC by nearly 20% over the average baseline and 2.8% over the best baseline.
It attains the top average rank across 57 ADBench datasets.
It provides complementary capability, performing strongly on datasets where many existing methods struggle.
It maintains overall superior performance compared to state-of-the-art methods.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

This implies that dependency-based methods can complement proximity-based ones for more complete anomaly coverage in tabular data.
The use of frozen models suggests potential for efficient deployment without per-dataset retraining.
Uncertainty awareness may reduce false positives from model limitations in noisy high-dimensional settings.

Load-bearing premise

Frozen PFNs trained on prior data can reliably estimate conditional dependencies in entirely new tabular datasets without any adaptation.

What would settle it

A controlled experiment on synthetic high-dimensional data with known dependency violations where uLEAD-TabPFN fails to achieve higher detection accuracy than standard methods like isolation forests or autoencoders.

Figures

Figures reproduced from arXiv: 2604.20255 by Craig Xie, Jiuyong Li, Jixue Liu, Lin Liu, Sha Lu, Stefan Peters, Thuc Duy Le.

**Figure 1.** Figure 1: The uLEAD-TabPFN framework consists of three components: Representative Context Set (RCS) construction, [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 2.** Figure 2: An illustrative example of uncertainty-aware scor [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Figure 3: Performance complementarity analysis of top [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: Sensitivity of uLEAD-TabPFN to latent dimension [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

**Figure 5.** Figure 5: UMAP visualization of the Wilt dataset in the orig [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗

**Figure 6.** Figure 6: Visulaization of top-5 and bottom-5 anomalies in both the original and latent feature spaces. [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗

**Figure 7.** Figure 7: The attribution of the orginal features to the anom [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗

read the original abstract

Anomaly detection in tabular data is challenging due to high dimensionality, complex feature dependencies, and heterogeneous noise. Many existing methods rely on proximity-based cues and may miss anomalies caused by violations of complex feature dependencies. Dependency-based anomaly detection provides a principled alternative by identifying anomalies as violations of dependencies among features. However, existing methods often struggle to model such dependencies robustly and to scale to high-dimensional data with complex dependency structures. To address these challenges, we propose uLEAD-TabPFN, a dependency-based anomaly detection framework built on Prior-Data Fitted Networks (PFNs). uLEAD-TabPFN identifies anomalies as violations of conditional dependencies in a learned latent space, leveraging frozen PFNs for dependency estimation. Combined with uncertainty-aware scoring, the proposed framework enables robust and scalable anomaly detection. Experiments on 57 tabular datasets from ADBench show that uLEAD-TabPFN achieves particularly strong performance in medium- and high-dimensional settings, where it attains the top average rank. On high-dimensional datasets, uLEAD-TabPFN improves the average ROC-AUC by nearly 20\% over the average baseline and by approximately 2.8\% over the best-performing baseline, while maintaining overall superior performance compared to state-of-the-art methods. Further analysis shows that uLEAD-TabPFN provides complementary anomaly detection capability, achieving strong performance on datasets where many existing methods struggle.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

uLEAD-TabPFN uses frozen TabPFNs for latent dependency estimation and uncertainty scoring, delivering clear gains on high-dimensional ADBench data while leaving the transfer assumption lightly checked.

read the letter

The paper's main contribution is a dependency-based anomaly detector that projects tabular data into latent space, uses a frozen PFN to estimate conditional dependencies, and scores high-uncertainty violations as anomalies. This is a straightforward new combination rather than an incremental tweak on existing proximity or reconstruction methods. It avoids per-dataset model fitting, which helps with scaling. The experiments on 57 ADBench datasets are the clearest strength: the method takes the top average rank, improves average ROC-AUC by nearly 20% over the average baseline on high-dimensional subsets, and edges the best baseline by about 2.8%. It also shows complementary performance on datasets where many other approaches fall short, which is practically useful. The results are reported cleanly enough to make the benchmark comparison credible. The soft spot is the frozen PFN itself. Because the model is pre-trained on synthetic priors and never adapted, its conditional estimates on new real tabular data rest on an untested transfer assumption. If the learned prior does not align with the target dataset's dependency structure, the violation scores risk conflating model mismatch with actual anomalies. The paper does not appear to include synthetic recovery tests, calibration plots, or direct checks on conditional accuracy, so the performance lift is encouraging but does not yet pin down why the mechanism succeeds. This work is aimed at the tabular anomaly detection community, especially anyone interested in PFN applications or dependency modeling in medium- and high-dimensional settings. A reader who wants benchmark numbers and a new baseline will get value from it. It deserves a serious referee because the experimental design is standard, the gains are measurable, and the idea is simple enough to reproduce and stress-test further. I would bring it to a reading group to examine the latent projection and scoring details. I would not cite it yet. It should go out for peer review.

Referee Report

2 major / 1 minor

Summary. The paper introduces uLEAD-TabPFN, a dependency-based anomaly detection framework that projects tabular data into a latent space, uses frozen TabPFN models to estimate conditional dependencies, and applies uncertainty-aware scoring to flag violations as anomalies. It reports results on 57 ADBench datasets, claiming the top average rank overall and particularly strong gains in medium- and high-dimensional regimes (nearly 20% average ROC-AUC improvement over baselines and 2.8% over the best baseline).

Significance. If the central performance claims hold, the work would demonstrate that pre-trained PFNs can be leveraged without adaptation or fine-tuning to provide scalable, complementary anomaly detection in high-dimensional tabular settings where proximity-based methods often fail.

major comments (2)

[Experimental Evaluation (high-dimensional results)] The headline ROC-AUC gains on high-dimensional ADBench subsets rest on the assumption that a frozen PFN pre-trained on synthetic data can produce reliable conditional dependency estimates for unseen real tabular distributions. The manuscript provides no calibration plots, synthetic recovery experiments, or direct comparison of estimated vs. ground-truth conditionals to validate this assumption, which is load-bearing for the dependency-violation scoring mechanism.
[Method (uLEAD-TabPFN framework)] The uncertainty-aware scoring is presented as the key enabler for robustness, yet the paper supplies no analysis of how uncertainty is computed from the PFN outputs or empirical evidence that high-uncertainty scores correlate with true anomalies rather than PFN model mismatch or noise.

minor comments (1)

[Abstract and Experiments] The abstract and results section would benefit from explicit mention of the number of runs, random seeds, and statistical significance tests supporting the reported average ranks and percentage improvements.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below and describe the revisions we will make to strengthen the manuscript.

read point-by-point responses

Referee: [Experimental Evaluation (high-dimensional results)] The headline ROC-AUC gains on high-dimensional ADBench subsets rest on the assumption that a frozen PFN pre-trained on synthetic data can produce reliable conditional dependency estimates for unseen real tabular distributions. The manuscript provides no calibration plots, synthetic recovery experiments, or direct comparison of estimated vs. ground-truth conditionals to validate this assumption, which is load-bearing for the dependency-violation scoring mechanism.

Authors: We agree that direct validation of the frozen PFN's conditional estimates on real distributions would provide stronger grounding for the dependency-violation mechanism. While the consistent top-rank performance across 57 real ADBench datasets (particularly the 2.8% gain over the best baseline in high dimensions) offers indirect empirical support for generalization, we acknowledge the value of explicit checks. In the revision we will add (i) synthetic recovery experiments that generate data with known ground-truth conditionals and quantify estimation fidelity, (ii) calibration plots of the PFN predictive distributions, and (iii) a brief comparison against available ground-truth conditionals where such information exists in the literature. These additions will appear in a new subsection of the experimental analysis. revision: yes
Referee: [Method (uLEAD-TabPFN framework)] The uncertainty-aware scoring is presented as the key enabler for robustness, yet the paper supplies no analysis of how uncertainty is computed from the PFN outputs or empirical evidence that high-uncertainty scores correlate with true anomalies rather than PFN model mismatch or noise.

Authors: The uncertainty score is derived from the predictive variance of the TabPFN posterior approximation (detailed in Section 3.2 of the manuscript). We recognize, however, that an explicit derivation and an empirical correlation study were not included. In the revised manuscript we will (i) expand the method section with a self-contained derivation of the uncertainty computation from the PFN output distribution and (ii) add an empirical analysis that reports Pearson/Spearman correlations between uncertainty scores and anomaly labels, together with visualizations on representative high-dimensional datasets. This will demonstrate that elevated uncertainty aligns with anomalies beyond mere model mismatch. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical claims rest on external benchmarks and frozen pre-trained models

full rationale

The paper introduces uLEAD-TabPFN as a framework that applies frozen, pre-trained TabPFN models (from prior independent work) to estimate conditional dependencies in new tabular data, then scores anomalies via uncertainty-aware violation detection. No equations or steps in the provided abstract or description reduce the performance claims to quantities fitted on the target ADBench datasets by construction. The reported ROC-AUC gains and rankings are obtained from direct experimental evaluation on 57 held-out datasets, without self-definitional loops, fitted-input predictions, or load-bearing self-citations that collapse the central result to its inputs. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based solely on the abstract, no explicit free parameters, axioms, or invented entities are described. The method assumes generalization of pre-trained PFNs to new tabular data (a domain assumption) and relies on the existence of conditional dependencies that can be captured in a latent space.

pith-pipeline@v0.9.0 · 5573 in / 1347 out tokens · 62098 ms · 2026-05-10T01:02:00.130410+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

61 extracted references · 18 canonical work pages · 2 internal anchors

[1]

Charu C Aggarwal. 2016. An introduction to outlier analysis. InOutlier analysis. Springer, 1–34

2016
[2]

Olivier Bachem, Mario Lucic, and Andreas Krause. 2018. Scalable k-means clustering via lightweight coresets. InProceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining. 1119–1127

2018
[3]

Liron Bergman and Yedid Hoshen. 2020. Classification-Based Anomaly Detection for General Data. InInternational Conference on Learning Representations. https: //openreview.net/forum?id=HJgExaVFwS

2020
[4]

Markus M Breunig, Hans-Peter Kriegel, Raymond T Ng, and Jö Sander. 2000. LOF: identifying density-based local outliers. InProceedings of the 2000 ACM SIGMOD International Conference on Management of Data. 93–104

2000
[5]

Leonie Buck et al. 2023. Anomaly Detection in Mixed High-dimensional Molecu- lar Data.Bioinformatics39, 8 (2023), btad501. doi:10.1093/bioinformatics/btad501

work page doi:10.1093/bioinformatics/btad501 2023
[6]

Varun Chandola, Arindam Banerjee, and Vipin Kumar. 2009. Anomaly Detection: A Survey.Comput. Surveys41, 3 (2009), 15:1–15:58. doi:10.1145/1541880.1541882

work page doi:10.1145/1541880.1541882 2009
[7]

Marie Chavent, Vanessa Kuentz-Simonet, Alexandre Labenne, and Jérôme Saracco. 2021. Factor Analysis of Mixed Data for Anomaly Detection.Sta- tistical Analysis and Data Mining: The ASA Data Science Journal14, 4 (2021), 395–410. doi:10.1002/sam.11585

work page doi:10.1002/sam.11585 2021
[8]

Mohamed A ElBahaay, Tamer Arafa, and Nermin AbdelBaki. 2023. Unsupervised Anomaly Detection for Throughput Enhancement in Mobile Networks. In2023 International Conference on Computer and Applications (ICCA). IEEE, 1–6

2023
[9]

Isabelle Fauconnier and Gentiane Haesbroeck. 2009. Influence function and robust regression.The Annals of Statistics37, 2 (2009), 917–947

2009
[10]

Dan Feldman and Michael Langberg. 2011. A unified framework for approximat- ing and clustering data. InProceedings of the forty-third annual ACM symposium on Theory of computing. 569–578

2011
[11]

Cheng Feng. 2024. PARs: Predicate-based Association Rules for Efficient and Accurate Anomaly Explanation. InProceedings of the 33rd ACM International Conference on Information and Knowledge Management (CIKM). 612–621

2024
[12]

Benjamin Feuer et al. 2024. FoMo-0D: A Foundation Model for Zero-shot Outlier Detection via Prior-data Fitted Networks.arXiv preprint arXiv:2409.05672(2024). Foundation PFN-based model for tabular outlier detection

work page arXiv 2024
[13]

Ben Feuer, Chinmay Hegde, and Niv Cohen. 2023. Scaling TabPFN: Sketching and Feature Selection for Tabular Prior-Data Fitted Networks.arXiv preprint arXiv:2311.10380(2023)

work page arXiv 2023
[14]

Markus Goldstein and Andreas Dengel. 2012. Histogram-based outlier score (HBOS): A fast unsupervised anomaly detection algorithm.KI-2012: Poster and Demo Track759 (2012), 9–16

2012
[15]

Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yves Bengio. 2014. Generative adversarial nets. InAdvances in Neural Information Processing Systems, Vol. 27

2014
[16]

Sachin Goyal, Aditi Raghunathan, Moksh Jain, Harsha Vardhan Simhadri, and Pra- teek Jain. 2020. DROCC: Deep Robust One-Class Classification. InInternational Conference on Machine Learning. doi:10.48550/arXiv.2002.12718

work page doi:10.48550/arxiv.2002.12718 2020
[17]

William H. Greene. 2012.Econometric Analysis. Pearson Education

2012
[18]

Siyi Han, Han Xiao, Yong Li, Chen Gong, Xueyan Zhou, Jian Pei, Lijun Wu, Ce Zhang, et al. 2022. ADBench: Anomaly Detection Benchmark. InAdvances in Neural Information Processing Systems, Vol. 35. 32142–32157

2022
[19]

Z He, X Xu, and S Deng. 2003. Discovering cluster-based local outliers.Pattern Recognition36, 12 (2003), 2701–2712. doi:10.1016/S0031-3203(03)00003-5

work page doi:10.1016/s0031-3203(03)00003-5 2003
[20]

Manuel Hirth and Enkelejda Kasneci. [n. d.]. Anomaly detection by estimating gradients of the tabular data distribution
[21]

Nina Hollmann, Samuel Müller, et al. 2025. Accurate Predictions on Small Data with a Tabular Prior-data Fitted Network.Nature627 (2025), 123–130. doi:10. 1038/s41586-024-08328-6 Introduces TabPFN and discusses its use for anomaly detection

2025
[22]

Hajar Homayouni, Hamed Aghayarzadeh, Indrakshi Ray, and Hossein Shirazi
[23]

In2024 IEEE International Conference on Data Mining Work- shops (ICDMW)

Anomaly Detection and Interpretation from Tabular Data Using Trans- former Architecture. In2024 IEEE International Conference on Data Mining Work- shops (ICDMW). IEEE, 657–664
[24]

Thong Huynh, Leonard Siu Tan, and Zeynep Akata. 2024. DAGnosis: Localized Identification of Data Inconsistencies using Structures. InProceedings of the 27th International Conference on Artificial Intelligence and Statistics (AISTATS)

2024
[25]

Brian Johnson. 2013. Wilt. UCI Machine Learning Repository. DOI: https://doi.org/10.24432/C5KS4M

work page doi:10.24432/c5ks4m 2013
[26]

Diederik P Kingma and Max Welling. 2013. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114(2013)

work page internal anchor Pith review Pith/arXiv arXiv 2013
[27]

Aleksandar Lazarevic and Vipin Kumar. 2005. Feature bagging for anomaly detection in web traffic. InProceedings of the 11th Pacific-Asia Conference on Advances in Knowledge Discovery and Data Mining. Springer, 371–382

2005
[28]

2025.Trustworthy Anomaly Detection for Smart Manufacturing

Zhong Li. 2025.Trustworthy Anomaly Detection for Smart Manufacturing. Ph. D. Dissertation. Leiden University

2025
[29]

Zhong Li and Matthijs van Leeuwen. 2023. Explainable contextual anomaly detection using quantile regression forests.Machine Learning Journal(2023)

2023
[30]

Zhao Li, Yue Zhao, Rongqi Cai, Mengdi Zhou, and Yongjun Su. 2022. ECOD: Un- supervised Outlier Detection Using Empirical Cumulative Distribution Functions. InInternational Conference on Learning Representations (ICLR). doi:10.48550/ arXiv.2201.00382

work page arXiv 2022
[31]

Zhao Li, Yue Zhao, Xiao Hu, Xiyang Chen, Xiyuan Deng, and Yongjun Su. 2020. COPOD: Copula-Based Outlier Detection. InIEEE International Conference on Data Mining (ICDM). doi:10.48550/arXiv.2009.09463

work page doi:10.48550/arxiv.2009.09463 2020
[32]

Fei Tony Liu, Kai Ming Ting, and Zhi-Hua Zhou. 2008. Isolation forest. InEighth IEEE International Conference on Data Mining. IEEE, 413–422. doi:10.1109/ICDM. 2008.17

work page doi:10.1109/icdm 2008
[33]

Victor Livernoche, Vineet Jain, Yashar Hezaveh, and Siamak Ravanbakhsh. 2024. On Diffusion Modeling for Anomaly Detection. InThe Twelfth International Conference on Learning Representations. doi:10.48550/arXiv.2305.18593 Accepted to ICLR 2024

work page internal anchor Pith review doi:10.48550/arxiv.2305.18593 2024
[34]

Sha Lu, Lin Liu, Jiuyong Li, Thuc Duy Le, and Jixue Liu. 2020. Lopad: A local pre- diction approach to anomaly detection. InPacific-Asia Conference on Knowledge Discovery and Data Mining. Springer, 660–673

2020
[35]

Sha Lu, Lin Liu, Kui Yu, Thuc Duy Le, Jixue Liu, and Jiuyong Li. 2025. Dependency- based anomaly detection: A general framework and comprehensive evaluation. Expert Systems with Applications(2025), 129249

2025
[36]

Kimberly T Mai, Toby Davies, and Lewis D Griffin. 2024. Understanding the limitations of self-supervised learning for tabular anomaly detection.Pattern Analysis and Applications27, 2 (2024), 61

2024
[37]

Rosario N Mantegna. 1999. Hierarchical structure in financial markets.The European Physical Journal B-Condensed Matter and Complex Systems11, 1 (1999), 193–197

1999
[38]

Samuel G Müller, Steven Adriaensen, et al. 2022. Prior-data fitted networks.arXiv preprint arXiv:2207.08914(2022)

work page arXiv 2022
[39]

Guansong Pang, Chunhua Shen, Longbing Cao, and Anton van den Hengel. 2021. Deep Learning for Anomaly Detection: A Review.Comput. Surveys54, 2 (2021), 38:1–38:38. doi:10.1145/3439950

work page doi:10.1145/3439950 2021
[40]

Tomás Pevný. 2016. Loda: Lightweight on-line detector of anomalies.Machine Learning102 (2016), 275–304

2016
[41]

Sridhar Ramaswamy, Raghu Rastogi, and Kyuseok Shim. 2000. Efficient algo- rithms for mining outliers from large data sets. InProceedings of the 2000 ACM SIGMOD International Conference on Management of Data. 427–438

2000
[42]

Jing Ren, Tao Tang, Hong Jia, Haytham Fayek, Xiaodong Li, Suyu Ma, Xiwei Xu, and Feng Xia. 2025. Foundation Models for Anomaly Detection: Vision and Challenges.arXiv preprint arXiv:2502.06537(2025)

work page arXiv 2025
[43]

Danilo Jimenez Rezende and Shakir Mohamed. 2015. Variational inference with normalizing flows. InInternational Conference on Machine Learning. PMLR, 1530– 1538

2015
[44]

Vandermeulen, Grégoire Montavon, Klaus-Robert Müller, and Marius Kloft

Lukas Ruff, Jacob Kauffmann, Robert A. Vandermeulen, Grégoire Montavon, Klaus-Robert Müller, and Marius Kloft. 2021. A Unifying Review of Deep and Shallow Anomaly Detection.Proc. IEEE109, 5 (2021), 756–795

2021
[45]

Lukas Ruff, Robert A Vandermeulen, Nico Göring, Vincent Deischinger, Konrad Binder, Elias Müller, and Marius Kloft. 2018. Deep one-class classification. In International Conference on Machine Learning. PMLR, 4393–4402

2018
[46]

Timur Sattarov, Marco Schreyer, and Damian Borth. 2025. Diffusion-Scheduled Denoising Autoencoders for Anomaly Detection in Tabular Data. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V

2025
[47]

Bernhard Schölkopf, Robert C Williamson, Alexander J Smola, John Shawe-Taylor, and John C Platt. 1999. Support vector method for novelty detection. InAdvances in Neural Information Processing Systems, Vol. 12

1999
[48]

Aviv Shenkar and Lior Wolf. 2022. One-Class Learning Towards General Anomaly Detection. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

2022
[49]

Mei-Ling Shyu, Shu-Ching Chen, Kanoksri Sarinnapakorn, and Liwei Chang. 2003. A novel anomaly detection scheme based on principal component classifier. In Conference acronym ’XX, June 03–05, 2018, Woodstock, NY Sha Lu, Jixue Liu, Stefan Peters, Thuc Duy Le, Lin Liu, Jiuyong Li, and Yongzheng Xie Proceedings of the IEEE Foundations and New Directions of Da...

2003
[50]

Hugo Thimonier, Fabrice Popineau, Arpad Rimmel, and Bich-Liên Doan. 2024. Be- yond individual input for deep anomaly detection on tabular data. InProceedings of the 41st International Conference on Machine Learning. 48097–48123

2024
[51]

Srujana Thudumu, Philip Branch, Jia Jin, and Andy Kenyon. 2020. A Compre- hensive Survey of Anomaly Detection Techniques for High Dimensional Data. Journal of Big Data7, 1 (2020), 26. doi:10.1186/s40537-020-00324-9

work page doi:10.1186/s40537-020-00324-9 2020
[52]

Che-Ping Tsai, Ganyu Teng, Phil Wallis, and Wei Ding. 2025. AnoLLM: Large Language Models for Tabular Anomaly Detection. InInternational Conference on Learning Representations (ICLR)

2025
[53]

Yuxin Wang, Botian Jiang, Yiran Guo, Quan Gan, David Wipf, Xuanjing Huang, and Xipeng Qiu. 2025. Prior-Fitted Networks Scale to Larger Datasets When Treated as Weak Learners.Proceedings of the 28th International Conference on Artificial Intelligence and Statistics (AISTATS)(2025)

2025
[54]

Yongzheng Xie, Hongyu Zhang, Bo Zhang, Muhammad Ali Babar, and Sha Lu
[55]

InInternational Conference on Service-Oriented Computing

Logdp: Combining dependency and proximity for log-based anomaly detection. InInternational Conference on Service-Oriented Computing. Springer, 708–716
[56]

Kui Xu et al. 2023. Diffusion-based Inverse Forensics for Anomaly Detection. In International Conference on Learning Representations

2023
[57]

Kui Xu et al . 2023. Scale Learning for Anomaly Detection. InInternational Conference on Machine Learning

2023
[58]

Jihun Yin, Yuan Qiao, Zhipeng Zhou, Xiyuan Wang, and Junchi Yang. 2024. MCM: Masked Cell Modeling for Anomaly Detection in Tabular Data. InInternational Conference on Learning Representations (ICLR)

2024
[59]

Xin Zhao, Xiao Leng, Luyu Wang, and et al. 2025. Efficient anomaly detection in tabular cybersecurity data using large language models.Scientific Reports15 (2025), 11772594

2025
[60]

Bo Zong, Qi Song, Martin Renqiang Min, Wei Cheng, Chunming Luan, and Ying Liu. 2018. Deep autoencoding gaussian mixture model for unsupervised anomaly detection. InInternational Conference on Learning Representations

2018
[61]

Tommaso Zoppi et al. 2024. Anomaly-based Error and Intrusion Detection in Tabular Data.Future Generation Computer Systems(2024). doi:10.1016/j.future. 2024.09.002 A Appendix: Datasets Details The dataset information used in the experiments is presented in Table 3. B Appendix: Baseline Methods We compare uLEAD-TabPFN against a total of 26 baseline methods ...

work page doi:10.1016/j.future 2024

[1] [1]

Charu C Aggarwal. 2016. An introduction to outlier analysis. InOutlier analysis. Springer, 1–34

2016

[2] [2]

Olivier Bachem, Mario Lucic, and Andreas Krause. 2018. Scalable k-means clustering via lightweight coresets. InProceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining. 1119–1127

2018

[3] [3]

Liron Bergman and Yedid Hoshen. 2020. Classification-Based Anomaly Detection for General Data. InInternational Conference on Learning Representations. https: //openreview.net/forum?id=HJgExaVFwS

2020

[4] [4]

Markus M Breunig, Hans-Peter Kriegel, Raymond T Ng, and Jö Sander. 2000. LOF: identifying density-based local outliers. InProceedings of the 2000 ACM SIGMOD International Conference on Management of Data. 93–104

2000

[5] [5]

Leonie Buck et al. 2023. Anomaly Detection in Mixed High-dimensional Molecu- lar Data.Bioinformatics39, 8 (2023), btad501. doi:10.1093/bioinformatics/btad501

work page doi:10.1093/bioinformatics/btad501 2023

[6] [6]

Varun Chandola, Arindam Banerjee, and Vipin Kumar. 2009. Anomaly Detection: A Survey.Comput. Surveys41, 3 (2009), 15:1–15:58. doi:10.1145/1541880.1541882

work page doi:10.1145/1541880.1541882 2009

[7] [7]

Marie Chavent, Vanessa Kuentz-Simonet, Alexandre Labenne, and Jérôme Saracco. 2021. Factor Analysis of Mixed Data for Anomaly Detection.Sta- tistical Analysis and Data Mining: The ASA Data Science Journal14, 4 (2021), 395–410. doi:10.1002/sam.11585

work page doi:10.1002/sam.11585 2021

[8] [8]

Mohamed A ElBahaay, Tamer Arafa, and Nermin AbdelBaki. 2023. Unsupervised Anomaly Detection for Throughput Enhancement in Mobile Networks. In2023 International Conference on Computer and Applications (ICCA). IEEE, 1–6

2023

[9] [9]

Isabelle Fauconnier and Gentiane Haesbroeck. 2009. Influence function and robust regression.The Annals of Statistics37, 2 (2009), 917–947

2009

[10] [10]

Dan Feldman and Michael Langberg. 2011. A unified framework for approximat- ing and clustering data. InProceedings of the forty-third annual ACM symposium on Theory of computing. 569–578

2011

[11] [11]

Cheng Feng. 2024. PARs: Predicate-based Association Rules for Efficient and Accurate Anomaly Explanation. InProceedings of the 33rd ACM International Conference on Information and Knowledge Management (CIKM). 612–621

2024

[12] [12]

Benjamin Feuer et al. 2024. FoMo-0D: A Foundation Model for Zero-shot Outlier Detection via Prior-data Fitted Networks.arXiv preprint arXiv:2409.05672(2024). Foundation PFN-based model for tabular outlier detection

work page arXiv 2024

[13] [13]

Ben Feuer, Chinmay Hegde, and Niv Cohen. 2023. Scaling TabPFN: Sketching and Feature Selection for Tabular Prior-Data Fitted Networks.arXiv preprint arXiv:2311.10380(2023)

work page arXiv 2023

[14] [14]

Markus Goldstein and Andreas Dengel. 2012. Histogram-based outlier score (HBOS): A fast unsupervised anomaly detection algorithm.KI-2012: Poster and Demo Track759 (2012), 9–16

2012

[15] [15]

Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yves Bengio. 2014. Generative adversarial nets. InAdvances in Neural Information Processing Systems, Vol. 27

2014

[16] [16]

Sachin Goyal, Aditi Raghunathan, Moksh Jain, Harsha Vardhan Simhadri, and Pra- teek Jain. 2020. DROCC: Deep Robust One-Class Classification. InInternational Conference on Machine Learning. doi:10.48550/arXiv.2002.12718

work page doi:10.48550/arxiv.2002.12718 2020

[17] [17]

William H. Greene. 2012.Econometric Analysis. Pearson Education

2012

[18] [18]

Siyi Han, Han Xiao, Yong Li, Chen Gong, Xueyan Zhou, Jian Pei, Lijun Wu, Ce Zhang, et al. 2022. ADBench: Anomaly Detection Benchmark. InAdvances in Neural Information Processing Systems, Vol. 35. 32142–32157

2022

[19] [19]

Z He, X Xu, and S Deng. 2003. Discovering cluster-based local outliers.Pattern Recognition36, 12 (2003), 2701–2712. doi:10.1016/S0031-3203(03)00003-5

work page doi:10.1016/s0031-3203(03)00003-5 2003

[20] [20]

Manuel Hirth and Enkelejda Kasneci. [n. d.]. Anomaly detection by estimating gradients of the tabular data distribution

[21] [21]

Nina Hollmann, Samuel Müller, et al. 2025. Accurate Predictions on Small Data with a Tabular Prior-data Fitted Network.Nature627 (2025), 123–130. doi:10. 1038/s41586-024-08328-6 Introduces TabPFN and discusses its use for anomaly detection

2025

[22] [22]

Hajar Homayouni, Hamed Aghayarzadeh, Indrakshi Ray, and Hossein Shirazi

[23] [23]

In2024 IEEE International Conference on Data Mining Work- shops (ICDMW)

Anomaly Detection and Interpretation from Tabular Data Using Trans- former Architecture. In2024 IEEE International Conference on Data Mining Work- shops (ICDMW). IEEE, 657–664

[24] [24]

Thong Huynh, Leonard Siu Tan, and Zeynep Akata. 2024. DAGnosis: Localized Identification of Data Inconsistencies using Structures. InProceedings of the 27th International Conference on Artificial Intelligence and Statistics (AISTATS)

2024

[25] [25]

Brian Johnson. 2013. Wilt. UCI Machine Learning Repository. DOI: https://doi.org/10.24432/C5KS4M

work page doi:10.24432/c5ks4m 2013

[26] [26]

Diederik P Kingma and Max Welling. 2013. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114(2013)

work page internal anchor Pith review Pith/arXiv arXiv 2013

[27] [27]

Aleksandar Lazarevic and Vipin Kumar. 2005. Feature bagging for anomaly detection in web traffic. InProceedings of the 11th Pacific-Asia Conference on Advances in Knowledge Discovery and Data Mining. Springer, 371–382

2005

[28] [28]

2025.Trustworthy Anomaly Detection for Smart Manufacturing

Zhong Li. 2025.Trustworthy Anomaly Detection for Smart Manufacturing. Ph. D. Dissertation. Leiden University

2025

[29] [29]

Zhong Li and Matthijs van Leeuwen. 2023. Explainable contextual anomaly detection using quantile regression forests.Machine Learning Journal(2023)

2023

[30] [30]

Zhao Li, Yue Zhao, Rongqi Cai, Mengdi Zhou, and Yongjun Su. 2022. ECOD: Un- supervised Outlier Detection Using Empirical Cumulative Distribution Functions. InInternational Conference on Learning Representations (ICLR). doi:10.48550/ arXiv.2201.00382

work page arXiv 2022

[31] [31]

Zhao Li, Yue Zhao, Xiao Hu, Xiyang Chen, Xiyuan Deng, and Yongjun Su. 2020. COPOD: Copula-Based Outlier Detection. InIEEE International Conference on Data Mining (ICDM). doi:10.48550/arXiv.2009.09463

work page doi:10.48550/arxiv.2009.09463 2020

[32] [32]

Fei Tony Liu, Kai Ming Ting, and Zhi-Hua Zhou. 2008. Isolation forest. InEighth IEEE International Conference on Data Mining. IEEE, 413–422. doi:10.1109/ICDM. 2008.17

work page doi:10.1109/icdm 2008

[33] [33]

Victor Livernoche, Vineet Jain, Yashar Hezaveh, and Siamak Ravanbakhsh. 2024. On Diffusion Modeling for Anomaly Detection. InThe Twelfth International Conference on Learning Representations. doi:10.48550/arXiv.2305.18593 Accepted to ICLR 2024

work page internal anchor Pith review doi:10.48550/arxiv.2305.18593 2024

[34] [34]

Sha Lu, Lin Liu, Jiuyong Li, Thuc Duy Le, and Jixue Liu. 2020. Lopad: A local pre- diction approach to anomaly detection. InPacific-Asia Conference on Knowledge Discovery and Data Mining. Springer, 660–673

2020

[35] [35]

Sha Lu, Lin Liu, Kui Yu, Thuc Duy Le, Jixue Liu, and Jiuyong Li. 2025. Dependency- based anomaly detection: A general framework and comprehensive evaluation. Expert Systems with Applications(2025), 129249

2025

[36] [36]

Kimberly T Mai, Toby Davies, and Lewis D Griffin. 2024. Understanding the limitations of self-supervised learning for tabular anomaly detection.Pattern Analysis and Applications27, 2 (2024), 61

2024

[37] [37]

Rosario N Mantegna. 1999. Hierarchical structure in financial markets.The European Physical Journal B-Condensed Matter and Complex Systems11, 1 (1999), 193–197

1999

[38] [38]

Samuel G Müller, Steven Adriaensen, et al. 2022. Prior-data fitted networks.arXiv preprint arXiv:2207.08914(2022)

work page arXiv 2022

[39] [39]

Guansong Pang, Chunhua Shen, Longbing Cao, and Anton van den Hengel. 2021. Deep Learning for Anomaly Detection: A Review.Comput. Surveys54, 2 (2021), 38:1–38:38. doi:10.1145/3439950

work page doi:10.1145/3439950 2021

[40] [40]

Tomás Pevný. 2016. Loda: Lightweight on-line detector of anomalies.Machine Learning102 (2016), 275–304

2016

[41] [41]

Sridhar Ramaswamy, Raghu Rastogi, and Kyuseok Shim. 2000. Efficient algo- rithms for mining outliers from large data sets. InProceedings of the 2000 ACM SIGMOD International Conference on Management of Data. 427–438

2000

[42] [42]

Jing Ren, Tao Tang, Hong Jia, Haytham Fayek, Xiaodong Li, Suyu Ma, Xiwei Xu, and Feng Xia. 2025. Foundation Models for Anomaly Detection: Vision and Challenges.arXiv preprint arXiv:2502.06537(2025)

work page arXiv 2025

[43] [43]

Danilo Jimenez Rezende and Shakir Mohamed. 2015. Variational inference with normalizing flows. InInternational Conference on Machine Learning. PMLR, 1530– 1538

2015

[44] [44]

Vandermeulen, Grégoire Montavon, Klaus-Robert Müller, and Marius Kloft

Lukas Ruff, Jacob Kauffmann, Robert A. Vandermeulen, Grégoire Montavon, Klaus-Robert Müller, and Marius Kloft. 2021. A Unifying Review of Deep and Shallow Anomaly Detection.Proc. IEEE109, 5 (2021), 756–795

2021

[45] [45]

Lukas Ruff, Robert A Vandermeulen, Nico Göring, Vincent Deischinger, Konrad Binder, Elias Müller, and Marius Kloft. 2018. Deep one-class classification. In International Conference on Machine Learning. PMLR, 4393–4402

2018

[46] [46]

Timur Sattarov, Marco Schreyer, and Damian Borth. 2025. Diffusion-Scheduled Denoising Autoencoders for Anomaly Detection in Tabular Data. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V

2025

[47] [47]

Bernhard Schölkopf, Robert C Williamson, Alexander J Smola, John Shawe-Taylor, and John C Platt. 1999. Support vector method for novelty detection. InAdvances in Neural Information Processing Systems, Vol. 12

1999

[48] [48]

Aviv Shenkar and Lior Wolf. 2022. One-Class Learning Towards General Anomaly Detection. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

2022

[49] [49]

Mei-Ling Shyu, Shu-Ching Chen, Kanoksri Sarinnapakorn, and Liwei Chang. 2003. A novel anomaly detection scheme based on principal component classifier. In Conference acronym ’XX, June 03–05, 2018, Woodstock, NY Sha Lu, Jixue Liu, Stefan Peters, Thuc Duy Le, Lin Liu, Jiuyong Li, and Yongzheng Xie Proceedings of the IEEE Foundations and New Directions of Da...

2003

[50] [50]

Hugo Thimonier, Fabrice Popineau, Arpad Rimmel, and Bich-Liên Doan. 2024. Be- yond individual input for deep anomaly detection on tabular data. InProceedings of the 41st International Conference on Machine Learning. 48097–48123

2024

[51] [51]

Srujana Thudumu, Philip Branch, Jia Jin, and Andy Kenyon. 2020. A Compre- hensive Survey of Anomaly Detection Techniques for High Dimensional Data. Journal of Big Data7, 1 (2020), 26. doi:10.1186/s40537-020-00324-9

work page doi:10.1186/s40537-020-00324-9 2020

[52] [52]

Che-Ping Tsai, Ganyu Teng, Phil Wallis, and Wei Ding. 2025. AnoLLM: Large Language Models for Tabular Anomaly Detection. InInternational Conference on Learning Representations (ICLR)

2025

[53] [53]

Yuxin Wang, Botian Jiang, Yiran Guo, Quan Gan, David Wipf, Xuanjing Huang, and Xipeng Qiu. 2025. Prior-Fitted Networks Scale to Larger Datasets When Treated as Weak Learners.Proceedings of the 28th International Conference on Artificial Intelligence and Statistics (AISTATS)(2025)

2025

[54] [54]

Yongzheng Xie, Hongyu Zhang, Bo Zhang, Muhammad Ali Babar, and Sha Lu

[55] [55]

InInternational Conference on Service-Oriented Computing

Logdp: Combining dependency and proximity for log-based anomaly detection. InInternational Conference on Service-Oriented Computing. Springer, 708–716

[56] [56]

Kui Xu et al. 2023. Diffusion-based Inverse Forensics for Anomaly Detection. In International Conference on Learning Representations

2023

[57] [57]

Kui Xu et al . 2023. Scale Learning for Anomaly Detection. InInternational Conference on Machine Learning

2023

[58] [58]

Jihun Yin, Yuan Qiao, Zhipeng Zhou, Xiyuan Wang, and Junchi Yang. 2024. MCM: Masked Cell Modeling for Anomaly Detection in Tabular Data. InInternational Conference on Learning Representations (ICLR)

2024

[59] [59]

Xin Zhao, Xiao Leng, Luyu Wang, and et al. 2025. Efficient anomaly detection in tabular cybersecurity data using large language models.Scientific Reports15 (2025), 11772594

2025

[60] [60]

Bo Zong, Qi Song, Martin Renqiang Min, Wei Cheng, Chunming Luan, and Ying Liu. 2018. Deep autoencoding gaussian mixture model for unsupervised anomaly detection. InInternational Conference on Learning Representations

2018

[61] [61]

Tommaso Zoppi et al. 2024. Anomaly-based Error and Intrusion Detection in Tabular Data.Future Generation Computer Systems(2024). doi:10.1016/j.future. 2024.09.002 A Appendix: Datasets Details The dataset information used in the experiments is presented in Table 3. B Appendix: Baseline Methods We compare uLEAD-TabPFN against a total of 26 baseline methods ...

work page doi:10.1016/j.future 2024