pith. sign in

arxiv: 2509.00931 · v3 · submitted 2025-08-31 · 📊 stat.ML · cs.LG

Semi-Supervised Bayesian GANs with Log-Signatures for Uncertainty-Aware Credit Card Fraud Detection

David Hirnschall This is my paper

Pith reviewed 2026-05-18 19:18 UTC · model grok-4.3

classification 📊 stat.ML cs.LG
keywords semi-supervised learningBayesian GANlog-signaturescredit card fraud detectiontime series classificationWasserstein lossuncertainty quantification
0
0 comments X

The pith

A semi-supervised Bayesian GAN with log-signatures and a new Wasserstein loss improves credit card fraud detection on time series with limited labels while quantifying uncertainty.

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

The paper develops a framework that extends conditional GANs to generate synthetic transaction sequences aligned with unlabeled real data. It adds Bayesian inference to produce full predictive distributions rather than single predictions, and uses log-signatures to encode irregular, variable-length sequences. A combined loss pulls generated samples toward real unlabeled ones while raising accuracy on the labeled subset. Evaluation on the BankSim simulator across different label ratios shows gains in both standard accuracy metrics and fraud-specific measures. Readers care because fraud systems routinely face scarce labels, irregular data streams, and high costs for wrong alerts.

Core claim

The authors claim that a conditional GAN augmented with Bayesian layers and log-signature features, trained under a novel Wasserstein distance loss that aligns generated samples with unlabeled data while maximizing labeled classification accuracy, yields consistent improvements in fraud detection performance on the BankSim dataset under varying amounts of supervision.

What carries the argument

The novel Wasserstein distance-based loss that aligns generated samples with real unlabeled data while maximizing classification accuracy on labeled data, inside a Bayesian conditional GAN that encodes transaction histories via log-signatures.

If this is right

  • Consistent gains over benchmarks in global statistical and fraud-specific metrics on BankSim under varying label proportions.
  • Effective encoding and classification of irregularly sampled, variable-length transaction sequences.
  • Predictive distributions that support uncertainty-aware decisions instead of point estimates.
  • Improved semi-supervised performance when labeled data are scarce.

Where Pith is reading between the lines

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

  • The same combination of generative alignment, log-signature encoding, and Bayesian uncertainty could transfer to other scarce-label time-series problems such as medical event detection.
  • Uncertainty estimates could be used to route borderline cases to human review and thereby lower costly false-positive interventions.
  • The framework might support online updates on streaming transaction data without full retraining.

Load-bearing premise

The Wasserstein loss can simultaneously align generated samples to real unlabeled transactions and raise accuracy on labeled ones, and log-signatures give robust encodings for irregular variable-length sequences.

What would settle it

On the BankSim dataset with low labeled-sample ratios, the method shows no improvement or worse performance than standard semi-supervised baselines in both global and domain-specific metrics.

Figures

Figures reproduced from arXiv: 2509.00931 by David Hirnschall.

Figure 1
Figure 1. Figure 1: Network architectures for discriminator and conditional generator. [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Expected Cost@K for K=0.5 for various amounts of labeled samples [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Average 90% uncertainty interval width over five unlabelings by outcome (TP/FP/TN/FN) across [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Predictive distributions for four uncertain predictions, one from each category (TP, FP, TN, FN). [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
read the original abstract

We present a novel deep generative semi-supervised framework for credit card fraud detection, formulated as time series classification task. As financial transaction data streams grow in scale and complexity, traditional methods often require large labeled datasets, struggle with time series of irregular sampling frequencies and varying sequence lengths. To address these challenges, we extend conditional Generative Adversarial Networks (GANs) for targeted data augmentation, integrate Bayesian inference to obtain predictive distributions and quantify uncertainty, and leverage log-signatures for robust feature encoding of transaction histories. We introduce a novel Wasserstein distance-based loss to align generated and real unlabeled samples while simultaneously maximizing classification accuracy on labeled data. Our approach is evaluated on the BankSim dataset, a widely used simulator for credit card transaction data, under varying proportions of labeled samples, demonstrating consistent improvements over benchmarks in both global statistical and domain-specific metrics. These findings highlight the effectiveness of GAN-driven semi-supervised learning with log-signatures for irregularly sampled time series and emphasize the importance of uncertainty-aware predictions.

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.

Referee Report

2 major / 2 minor

Summary. The paper presents a semi-supervised Bayesian GAN framework for credit card fraud detection formulated as time series classification. It extends conditional GANs for data augmentation, incorporates Bayesian inference for uncertainty quantification, and uses log-signatures for encoding irregularly sampled and variable-length transaction sequences. A novel Wasserstein distance-based loss is introduced to align generated samples with real unlabeled data while maximizing classification accuracy on labeled data. The approach is evaluated on the BankSim dataset under varying proportions of labeled samples, claiming consistent improvements over benchmarks in global statistical and domain-specific metrics.

Significance. If the results hold and the dual-role loss is properly formulated with supporting ablations, this could advance semi-supervised methods for irregular financial time series by integrating generative augmentation, signature-based encoding, and Bayesian uncertainty quantification, potentially improving performance in data-scarce fraud detection scenarios.

major comments (2)
  1. [Abstract] Abstract: the claim of 'consistent improvements over benchmarks in both global statistical and domain-specific metrics' is asserted without any quantitative results, baseline details, statistical significance tests, or description of how irregular sampling and variable lengths are handled; this is load-bearing for the central empirical claim.
  2. [Methods] Methods (loss formulation): the novel Wasserstein distance-based loss is asserted to simultaneously align generated samples with real unlabeled data and maximize classification accuracy on labeled data, but no explicit combined objective, weighting schedule, or ablation isolating the Wasserstein term's contribution is provided; without this the improvements cannot be confidently attributed to the claimed mechanism.
minor comments (2)
  1. [Experiments] Experiments: clarify the exact procedure for incorporating log-signatures into the GAN discriminator or classifier to handle variable-length sequences.
  2. [Methods] Notation: ensure consistent use of symbols for the Wasserstein loss term and Bayesian predictive distribution across equations.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The comments highlight opportunities to strengthen the abstract and clarify the loss formulation. We address each point below and will revise the manuscript to improve transparency and rigor.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim of 'consistent improvements over benchmarks in both global statistical and domain-specific metrics' is asserted without any quantitative results, baseline details, statistical significance tests, or description of how irregular sampling and variable lengths are handled; this is load-bearing for the central empirical claim.

    Authors: We agree that the abstract would benefit from greater specificity to support the central claim. In the revised version we will incorporate key quantitative results (e.g., average AUC and F1 improvements across label proportions on BankSim), name the primary baselines, explicitly state that log-signatures are used to encode irregular sampling and variable-length sequences, and report that metrics are averaged over multiple independent runs with standard deviations. Full experimental details and statistical comparisons remain in the Experiments section. revision: yes

  2. Referee: [Methods] Methods (loss formulation): the novel Wasserstein distance-based loss is asserted to simultaneously align generated samples with real unlabeled data and maximize classification accuracy on labeled data, but no explicit combined objective, weighting schedule, or ablation isolating the Wasserstein term's contribution is provided; without this the improvements cannot be confidently attributed to the claimed mechanism.

    Authors: We acknowledge that the current presentation of the loss could be more explicit. We will add the full mathematical form of the combined objective, including the weighting coefficients between the Wasserstein alignment term and the supervised classification term, and describe the schedule used to balance these terms during training. We will also include a dedicated ablation study that removes or varies the Wasserstein component while keeping other elements fixed, allowing direct attribution of performance gains to this term. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical claims rest on external benchmarks

full rationale

The paper presents an empirical semi-supervised Bayesian GAN framework that augments conditional GANs with Bayesian inference, log-signature encoding, and a novel Wasserstein loss for aligning unlabeled data while improving labeled classification. Evaluation occurs on the BankSim simulator across varying label proportions, with reported gains over benchmarks in statistical and domain metrics. No derivation chain, equation, or self-citation reduces a central prediction or uniqueness claim to its own fitted inputs or prior author work by construction. The approach is self-contained against external data and comparisons, with no load-bearing self-referential steps.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Only the abstract is available, so specific free parameters, axioms, and invented entities cannot be extracted in detail. The framework implicitly relies on standard GAN convergence assumptions and properties of the Wasserstein metric.

axioms (1)
  • standard math Standard assumptions on GAN training convergence and properties of the Wasserstein distance
    Invoked by the use of a Wasserstein-based loss but not stated or justified in the abstract.

pith-pipeline@v0.9.0 · 5700 in / 1280 out tokens · 50172 ms · 2026-05-18T19:18:33.426721+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

59 extracted references · 59 canonical work pages · 1 internal anchor

  1. [1]

    Credit card fraud detection: A realistic modeling and a novel learning strategy.IEEE Transactions on Neural Networks and Learning Systems, 29(8):3784–3797, 2018

    Andrea Dal Pozzolo, Giacomo Boracchi, Olivier Caelen, Cesare Alippi, and Gianluca Bontempi. Credit card fraud detection: A realistic modeling and a novel learning strategy.IEEE Transactions on Neural Networks and Learning Systems, 29(8):3784–3797, 2018. doi: 10.1109/TNNLS.2017.2736643

    work page doi:10.1109/tnnls.2017.2736643 2018

  2. [2]

    Sequence classification for credit-card fraud detection.Expert Systems with Applications, 100:234–245, 2018

    Johannes Jurgovsky, Michael Granitzer, Klaus Ziegler, Sylvie Calabretto, Philippe Portier, Laurent He-Guelton, and Olivier Caelen. Sequence classification for credit-card fraud detection.Expert Systems with Applications, 100:234–245, 2018. doi: 10.1016/j.eswa.2018.01.037

    work page doi:10.1016/j.eswa.2018.01.037 2018

  3. [3]

    Almazroi and Nasir Ayub

    Abdullah A. Almazroi and Nasir Ayub. Online payment fraud detection model using machine learning techniques.IEEE Access, 11:137188–137203, 2023. doi: 10.1109/ACCESS.2023.3339226

    work page doi:10.1109/access.2023.3339226 2023

  4. [4]

    Mienye and Nick Jere

    Ishmael D. Mienye and Nick Jere. Deep learning for credit card fraud detection: A review of algorithms, challenges, and solutions.IEEE Access, 12:96893–96910, 2024. doi: 10.1109/ACCESS.2024.3426955

    work page doi:10.1109/access.2024.3426955 2024

  5. [5]

    Using generative adversarial networks for improving classification effectiveness in credit card fraud detection.Information Sciences, 479:448–455, 2019

    Ugo Fiore, Alfredo De Santis, Francesca Perla, Paolo Zanetti, and Francesco Palmieri. Using generative adversarial networks for improving classification effectiveness in credit card fraud detection.Information Sciences, 479:448–455, 2019

    work page 2019

  6. [6]

    Enhancing financial fraud detection through ad- dressing class imbalance using hybrid smote-gan techniques.International Journal of Financial Studies, 11(3):110, 2023

    Pek Cher Chew, Ying Yang, and Byoung-Gyu Lee. Enhancing financial fraud detection through ad- dressing class imbalance using hybrid smote-gan techniques.International Journal of Financial Studies, 11(3):110, 2023. doi: 10.3390/ijfs11030110

    work page doi:10.3390/ijfs11030110 2023

  7. [7]

    Bellotti

    Zhipeng Zhao, Tian Cui, Shuai Ding, Jing Li, and Anthony G. Bellotti. Resampling techniques study on class imbalance problem in credit risk prediction.Mathematics, 12(5):701, 2024. doi: 10.3390/ math12050701

    work page 2024

  8. [8]

    Raksha: A trusted explainable lstm model to classify fraud patterns on credit card transactions.Mathematics, 11(8):1901, 2023

    Jay Raval et al. Raksha: A trusted explainable lstm model to classify fraud patterns on credit card transactions.Mathematics, 11(8):1901, 2023. doi: 10.3390/math11081901

    work page doi:10.3390/math11081901 1901

  9. [9]

    Enhanced credit card fraud detection based on attention mechanism and lstm deep model.Journal of Big Data, 8(1):1–21, 2021

    Imane Benchaji, Soukaina Douzi, Brahim El Ouahidi, and Jamila Jaafari. Enhanced credit card fraud detection based on attention mechanism and lstm deep model.Journal of Big Data, 8(1):1–21, 2021. doi: 10.1186/s40537-021-00541-8

    work page doi:10.1186/s40537-021-00541-8 2021

  10. [10]

    Feature engineering strategies for credit card fraud detection.Expert Systems with Applications, 51:134–142,

    Alejandro Correa Bahnsen, Djamila Aouada, Aleksandar Stojanovic, and Bj¨ orn Ottersten. Feature engineering strategies for credit card fraud detection.Expert Systems with Applications, 51:134–142,

    work page

  11. [11]

    doi: https://doi.org/10.1016/j.eswa.2015.12.030

    work page doi:10.1016/j.eswa.2015.12.030 2015

  12. [12]

    A generalised signature method for multivariate time series feature extraction, 2020

    James Morrill, Augustin Fermanian, Patrick Kidger, and Terry Lyons. A generalised signature method for multivariate time series feature extraction, 2020. 18 Semi-Supervised Bayesian GANs with Log-Signatures for Uncertainty-Aware Credit Card Fraud DetectionA Preprint

    work page 2020

  13. [13]

    Time series classification: A review of algorithms and implementations

    J´ erome Faouzi. Time series classification: A review of algorithms and implementations. InTime Series Analysis – Recent Advances, New Perspectives and Applications. IntechOpen, 2024. doi: 10.5772/ intechopen.1004810

    work page 2024

  14. [14]

    Unsupervised and semi-supervised learning with categorical generative ad- versarial networks, 2015

    Jost Tobias Springenberg. Unsupervised and semi-supervised learning with categorical generative ad- versarial networks, 2015

    work page 2015

  15. [15]

    Van Engelen and Holger H

    Jesper E. Van Engelen and Holger H. Hoos. A survey on semi-supervised learning.Machine Learning, 109(2):373–440, 2020. doi: 10.1007/s10994-019-05855-6

    work page doi:10.1007/s10994-019-05855-6 2020

  16. [16]

    Banksim: A bank payments simulator for fraud detection research

    Edgar Alonso Lopez-Rojas and Sam Axelsson. Banksim: A bank payments simulator for fraud detection research. InProceedings of the 26th European Modeling and Simulation Symposium (EMSS), pages 144– 152, Bordeaux, France, 2014

    work page 2014

  17. [17]

    Label propagation for deep semi- supervised learning

    Ahmet Iscen, Giorgos Tolias, Yannis Avrithis, and Ondrej Chum. Label propagation for deep semi- supervised learning. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 5070–5079, Long Beach, CA, USA, 2019

    work page 2019

  18. [18]

    Mixmatch: A holistic approach to semi-supervised learning

    David Berthelot, Nicholas Carlini, Ian Goodfellow, Nicolas Papernot, Avital Oliver, and Colin Raf- fel. Mixmatch: A holistic approach to semi-supervised learning. InAdvances in Neural Information Processing Systems, volume 33, pages 5049–5059, 2019. doi: 10.5555/3454287.3454741

    work page doi:10.5555/3454287.3454741 2019

  19. [19]

    Deep semi-supervised learning for time-series classification

    Julian Goschenhofer. Deep semi-supervised learning for time-series classification. InDeep Learning Applications, volume 4, pages 361–384. Springer, 2022. doi: 10.1007/978-981-19-6153-3 15

    work page doi:10.1007/978-981-19-6153-3 2022

  20. [20]

    Self-supervised learning for semi-supervised time series classification

    Saad Jawed, Josif Grabocka, and Lars Schmidt-Thieme. Self-supervised learning for semi-supervised time series classification. InAdvances in Knowledge Discovery and Data Mining, pages 499–511, Sin- gapore, 2020. doi: 10.1007/978-3-030-47426-3 39

    work page doi:10.1007/978-3-030-47426-3 2020

  21. [21]

    Milad Rezagholiradeh and Morteza A. Haidar. Reg-gan: Semi-supervised learning based on generative adversarial networks for regression. InProceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 2806–2810, Calgary, AB, Canada, 2018. doi: 10.1109/ ICASSP.2018.8462534

    work page arXiv 2018

  22. [22]

    Improved techniques for training gans

    Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved techniques for training gans. InProceedings of the 30th International Conference on Neural Information Processing Systems, pages 2234–2242, Barcelona, Spain, 2016. doi: 10.5555/3157096.3157346

    work page doi:10.5555/3157096.3157346 2016

  23. [23]

    How does gan-based semi-supervised learning work?, 2020

    Xiaoyu Liu and Xing Xiang. How does gan-based semi-supervised learning work?, 2020

    work page 2020

  24. [24]

    Triple generative adversarial networks

    Chongxuan Li, Kun Xu, Jun Zhu, Jianfeng Liu, and Bo Zhang. Triple generative adversarial networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(12):9629–9640, 2022. doi: 10. 1109/TPAMI.2021.3127558

    work page arXiv 2022

  25. [25]

    Segan: Semi-supervised learning approach for missing data imputation, 2024

    Xiaomin Pan et al. Segan: Semi-supervised learning approach for missing data imputation, 2024

    work page 2024

  26. [26]

    Semi-supervised generative ad- versarial networks with spatial coevolution for enhanced image generation and classification.Applied Soft Computing, 148, 2023

    Juan Toutouh, Sreya Nalluru, Erik Hemberg, and Una-May O’Reilly. Semi-supervised generative ad- versarial networks with spatial coevolution for enhanced image generation and classification.Applied Soft Computing, 148, 2023. doi: 10.1016/j.asoc.2023.110890

    work page doi:10.1016/j.asoc.2023.110890 2023

  27. [27]

    Gan-based anomaly detection tailored for classifiers.Mathematics, 12(10):1439, 2024

    Luk´ aˇ s Kr´ alik, Matej Kontˇ sek, OndrejˇSkvarek, and Miroslav Klimo. Gan-based anomaly detection tailored for classifiers.Mathematics, 12(10):1439, 2024. doi: 10.3390/math12101439

    work page doi:10.3390/math12101439 2024

  28. [28]

    Iterated integrals and exponential homomorphisms.Proceedings of the London Math- ematical Society, 4(1):502–512, 1954

    Kuo-Tsai Chen. Iterated integrals and exponential homomorphisms.Proceedings of the London Math- ematical Society, 4(1):502–512, 1954. doi: 10.1112/plms/s3-4.1.502

    work page doi:10.1112/plms/s3-4.1.502 1954

  29. [29]

    Terry J. Lyons. Differential equations driven by rough signals.Revista Matem´ atica Iberoamericana, 14 (2):215–310, 1998. URLhttp://eudml.org/doc/39555

    work page 1998

  30. [30]

    Path signatures for feature extraction: An introduction to the mathematics under- pinning an efficient machine learning technique, 2025

    Sebastian Sturm. Path signatures for feature extraction: An introduction to the mathematics under- pinning an efficient machine learning technique, 2025

    work page 2025

  31. [31]

    Numerical method for model-free pricing of exotic derivatives in discrete time using rough path signatures.Applied Mathematical Finance, 26(6): 583–597, 2020

    Terry Lyons, Sohrab Nejad, and Ignacio Perez Arribas. Numerical method for model-free pricing of exotic derivatives in discrete time using rough path signatures.Applied Mathematical Finance, 26(6): 583–597, 2020. doi: 10.1080/1350486x.2020.1726784

    work page doi:10.1080/1350486x.2020.1726784 2020

  32. [32]

    Developing the path signature methodology and its application to landmark-based human action recognition

    Weixin Yang, Terry Lyons, Hao Ni, Cordelia Schmid, and Li Jin. Developing the path signature methodology and its application to landmark-based human action recognition. InStochastic Analysis, Filtering, and Stochastic Optimization, pages 431–464. Springer, 2022. doi: 10.1007/978-3-030-98519-6 19. 19 Semi-Supervised Bayesian GANs with Log-Signatures for Un...

    work page doi:10.1007/978-3-030-98519-6 2022

  33. [33]

    Adaptive global gesture paths and signature features for skeleton-based gesture recognition

    Dan Shi, Xin Zhang, Jing Cheng, Tian Xiong, and Hao Ni. Adaptive global gesture paths and signature features for skeleton-based gesture recognition. InPattern Recognition, pages 278–292. Springer Nature Switzerland, Cham, 2025. doi: 10.1007/978-3-031-78354-8 18

    work page doi:10.1007/978-3-031-78354-8 2025

  34. [34]

    The path to a goal: Understanding soccer possessions via path signatures, 2025

    David Hirnschall and Robert Bajons. The path to a goal: Understanding soccer possessions via path signatures, 2025

    work page 2025

  35. [35]

    A data-driven market simulator for small data environments, 2020

    Hans Buehler, Blanka Horvath, Terry Lyons, Ignacio Perez Arribas, and Ben Wood. A data-driven market simulator for small data environments, 2020. SSRN preprint 3632431

    work page 2020

  36. [36]

    Learning stochastic differential equations using rnn with log signature features, 2019

    Shujian Liao, Terry Lyons, Weixin Yang, and Hao Ni. Learning stochastic differential equations using rnn with log signature features, 2019

    work page 2019

  37. [37]

    Sig-wasserstein gans for time series generation

    Hao Ni, Ignacio Perez Arribas, Terry Lyons, and Weixin Yang. Sig-wasserstein gans for time series generation. InProceedings of the 2nd ACM International Conference on AI in Finance, New York, NY, USA, 2021. doi: 10.1145/3490354.3494393

    work page doi:10.1145/3490354.3494393 2021

  38. [38]

    Applications of signature methods to market anomaly detection, 2022

    Ekin Akyildirim, Marco Gambara, Josef Teichmann, and Shenglong Zhou. Applications of signature methods to market anomaly detection, 2022

    work page 2022

  39. [39]

    Sigformer: Signature transformers for deep hedging

    Alan Tong, Tri Nguyen-Tang, Dongsu Lee, Trung Minh Tran, and Jaehyuk Choi. Sigformer: Signature transformers for deep hedging. InProceedings of the 4th ACM International Conference on AI in Finance (ICAIF), pages 124–132, Brooklyn, NY, USA, 2023. doi: 10.1145/3604237.3626841

    work page doi:10.1145/3604237.3626841 2023

  40. [40]

    Rough transformers: Lightweight continuous-time sequence modelling with path signatures, 2024

    Fernando Moreno-Pino, Adri´ an Arroyo, Henry Waldon, Xinzhan Dong, and ´Alvaro Cartea. Rough transformers: Lightweight continuous-time sequence modelling with path signatures, 2024

    work page 2024

  41. [41]

    doi: 10.1145/3422622

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial networks.Communications of the ACM, 63(11): 139–144, 2020. doi: 10.1145/3422622

    work page doi:10.1145/3422622 2020

  42. [42]

    Wasserstein generative adversarial networks

    Martin Arjovsky, Soumith Chintala, and L´ eon Bottou. Wasserstein generative adversarial networks. In Proceedings of the 34th International Conference on Machine Learning (ICML), pages 214–223, Sydney, Australia, 2017

    work page 2017

  43. [43]

    Improved Training of Wasserstein GANs

    Ishaan Gulrajani, Faruk Ahmed, Martin Arjovsky, Vincent Dumoulin, and Aaron Courville. Improved training of wasserstein gans. InAdvances in Neural Information Processing Systems, volume 30, 2017. doi: 10.48550/arXiv.1704.00028

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1704.00028 2017

  44. [44]

    Bayesian gan

    Yusuf Saatci and Andrew Gordon Wilson. Bayesian gan. InProceedings of the 31st International Conference on Neural Information Processing Systems, volume 30, pages 3625—-3634, Red Hook, NY, USA, 2017. Curran Associates Inc. doi: 10.5555/3294996.3295120

    work page doi:10.5555/3294996.3295120 2017

  45. [45]

    Stochastic gradient hamiltonian monte carlo

    Tianqi Chen, Emily Fox, and Carlos Guestrin. Stochastic gradient hamiltonian monte carlo. InPro- ceedings of the 31st International Conference on Machine Learning, pages 1683–1691, Beijing, China,

    work page

  46. [46]

    doi: 10.5555/3044805.3045080

    work page doi:10.5555/3044805.3045080

  47. [47]

    Holmes, and Stephen G

    Pier Giovanni Bissiri, Chris C. Holmes, and Stephen G. Walker. A general framework for updating belief distributions.Journal of the Royal Statistical Society: Series B, 78(5):1103–1130, 2016. URL https://www.jstor.org/stable/44682909

    work page arXiv 2016

  48. [48]

    Uniqueness for the signature of a path of bounded variation and the reduced path group.Annals of Mathematics, 171(1):109–167, 2010

    Ben Hambly and Terry Lyons. Uniqueness for the signature of a path of bounded variation and the reduced path group.Annals of Mathematics, 171(1):109–167, 2010. doi: 10.4007/annals.2010.171.109

    work page doi:10.4007/annals.2010.171.109 2010

  49. [49]

    A primer on the signature method in machine learning, 2016

    Ilya Chevyrev and Andrey Kormilitzin. A primer on the signature method in machine learning, 2016

    work page 2016

  50. [50]

    Sig-wasserstein gans for conditional time series generation.Mathematical Finance, 34(2):622–670, 2024

    Shujian Liao, Terry Lyons, Hao Ni, Weixin Yang, Cordelia Schmid, and Li Jin. Sig-wasserstein gans for conditional time series generation.Mathematical Finance, 34(2):622–670, 2024. doi: 10.1111/mafi.12423

    work page doi:10.1111/mafi.12423 2024

  51. [51]

    Understanding the difficulty of training deep feedforward neural networks

    Xavier Glorot and Yoshua Bengio. Understanding the difficulty of training deep feedforward neural networks. InProceedings of the 13th International Conference on Artificial Intelligence and Statistics (AISTATS), volume 9, pages 249–256, Sardinia, Italy, 2010. PMLR. URLhttps://proceedings.mlr. press/v9/glorot10a.html

    work page 2010

  52. [52]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium. InProceedings of the 31st International Conference on Neural Information Processing Systems, volume 30, page 6629–6640, Red Hook, NY, USA, 2017. Curran Associates Inc. URLhttps://dl.a...

    work page 2017

  53. [53]

    An experimental study with imbalanced classification approaches for credit card fraud detection.IEEE Access, 7:93010–93022, 2019

    Saad Makki et al. An experimental study with imbalanced classification approaches for credit card fraud detection.IEEE Access, 7:93010–93022, 2019. doi: 10.1109/ACCESS.2019.2927266. 20 Semi-Supervised Bayesian GANs with Log-Signatures for Uncertainty-Aware Credit Card Fraud DetectionA Preprint

    work page doi:10.1109/access.2019.2927266 2019

  54. [54]

    Scikit-learn: Machine learning in python.Journal of Machine Learning Research, 12:2825–2830, 2011

    Fabian Pedregosa, Ga¨ el Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Weiss, Vincent Dubourg, Jake Vanderplas, Alexandre Passos, David Cournapeau, Matthieu Brucher, Matthieu Perrot, and Edouard Duchesnay. Scikit-learn: Machine learning in python.Journal of Machine Learning Resear...

    work page 2011

  55. [55]

    Learning long-term dependencies in irregularly-sampled time series, 2020

    Mathias Lechner and Ramin Hasani. Learning long-term dependencies in irregularly-sampled time series, 2020. URLhttps://arxiv.org/abs/2006.04418

    work page arXiv 2020

  56. [56]

    Raffel, Ekin D

    Avital Oliver, Augustus Odena, Colin A. Raffel, Ekin D. Cubuk, and Ian Goodfellow. Realistic evaluation of deep semi-supervised learning algorithms, 2018

    work page 2018

  57. [57]

    Data leakage and deceptive performance: A critical examination of credit card fraud detection methodologies, 2025

    Khizar Hayat and Bastien Magnier. Data leakage and deceptive performance: A critical examination of credit card fraud detection methodologies, 2025

    work page 2025

  58. [58]

    Manning, Prabhakar Raghavan, and Hinrich Sch¨ utze.Introduction to Information Retrieval

    Christopher D. Manning, Prabhakar Raghavan, and Hinrich Sch¨ utze.Introduction to Information Retrieval. Cambridge University Press, Cambridge, UK, 2008

    work page 2008

  59. [59]

    The foundations of cost-sensitive learning

    Charles Elkan. The foundations of cost-sensitive learning. InProceedings of the 17th International Joint Conference on Artificial Intelligence (IJCAI), pages 973–978, San Francisco, CA, USA, 2001. Morgan Kaufmann. doi: https://dl.acm.org/doi/10.5555/1642194.1642224. 21

    work page doi:10.5555/1642194.1642224 2001