Pith. sign in

REVIEW 5 major objections 4 minor 27 references

WBHT: A Generative Attention Architecture for Detecting Black Hole Anomalies in Backbone Networks

T0 review · 5 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A Wasserstein GAN whose LSTM-plus-attention encoder and ConvLSTM generator reconstruct normal traffic catches silent packet-dropping black hole anomalies at F1 0.9250, beating all twelve baselines on real ISP data.

desk verdict Plausible incremental GAN+attention benchmark for black hole anomalies, but the missing anomaly threshold makes the headline F1 comparison underdetermined. read the letter →

arxiv 2507.20373 v1 pith:FEESLUMW submitted 2025-07-27 cs.LG cs.AI

Kiymet Kaya , Elif Ak , Sule Gunduz Oguducu This is my paper · ORCID
classification cs.LGcs.AI
keywords blackholeanomalydetectionWassersteinGANmulti-headself-attentionLSTMtransformergenerativeadversarialnetworkssemi-supervisedbackbone
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that a hybrid generative-sequential-attention architecture, called WBHT, can detect black hole anomalies (network faults that silently drop packets without issuing alarms) better than either generative or transformer methods alone. On a real ISP dataset whose training portion contains exclusively normal traffic, WBHT reaches an F1 score of 0.9250, beating twelve baselines, with the closest competitor f-AnoGAN at 0.9099. The practical stakes are that black holes cause connectivity loss and financial damage precisely because they go undetected, and this scheme needs no labeled attack examples, only clean normal traffic for training. The paper argues that the Wasserstein loss gives stable training, the LSTM-plus-attention encoder supplies the inverse mapping WGANs lack, and attention refines the latent representation so that the reconstruction residual separates anomalies cleanly.

What carries the argument

The load-bearing object is the trained triple of encoder, generator, and discriminator. The encoder $E$ (stacked LSTM layers plus multi-head self-attention) maps input traffic $x$ to a latent code $z$; the ConvLSTM generator $G$ reconstructs $x$ from $z$; the Wasserstein discriminator $D$ provides both the adversarial loss and an intermediate feature layer $f(\cdot)$. Inference scores a sequence by $\mathcal{L} = \frac{1}{n}\sqrt{\sum_{t}(x - G(E(x)))^2} + \frac{k}{n_d}\sqrt{\sum_{t}(f(x) - f(G(E(x))))^2}$, a reconstruction residual plus a weighted feature-matching residual inside $D$'s latent space. The Wasserstein distance, enforced through weight-clipping's Lipschitz constraint, replaces Jensen-Shannon divergence to keep training stable, and the separately trained encoder (with $G$ and $D$ frozen) provides the inverse mapping $x \to z$ that a plain WGAN lacks, making inference fast. Attention is what lets the encoder weight which time steps matter, which the paper ties to catching bursty, short-duration black hole events.

What would settle it

Run the exact architecture and training recipe on a public network-traffic benchmark with verified-clean normal training windows and labeled black hole events, under several random seeds: if the reported margin over f-AnoGAN (roughly 1.5 F1 points) collapses, inverts, or falls inside run-to-run variance, the central claim fails. A second check is to audit the training portion of the proprietary dataset for leaked anomaly labels: if BH events appear in the normal-only split, the reconstruction-plus-discriminator score would be learning anomalies as normal, and the reported separation would be an artifact of label leakage.

Watch

Extended reading notes

Core claim

The paper's central claim, stated on its own terms, is that WBHT successfully integrates WGAN for stable training, LSTM-based encoding for sequential learning, and Multi-Head Attention for fine-grained feature extraction, allowing it to outperform all baseline models in black hole detection. Reported on real ISP data with BH labels used only at test time, the model achieves F1 0.9250, detection rate 0.9532, false-alarm rate 0.0780, and accuracy 0.9322, improving F1 over existing models by 1.65% to 58.76%. The ablation study shows Wasserstein-trained variants beating vanilla-GAN variants across every encoder-generator pairing, with the winning combination being a ConvLSTM generator, an LSTM-plus-multi-head-attention encoder, and a Wasserstein discriminator. The model learns the distribution of normal traffic and flags sequences whose reconstruction error and discriminator feature residual deviate from it, which the paper argues is why it localizes the short, bursty intervals typical of black hole events while transformer-only models fail to.

Load-bearing premise

The load-bearing premise is that the training data contains exclusively normal traffic and no black hole events, with anomaly labels used only when scoring the test set, and because the ISP dataset is proprietary and unreleased, that cleanliness cannot be audited from outside.

Editorial extensions

If this is right

  • If the claim holds, operators can deploy black hole detection with no labeled anomalies at all: only a training window verified to contain normal traffic is required, which matters because black hole events are rare, unannounced, and almost never labeled.
  • The ablation table implies that the Wasserstein loss, not any single layer type, carries most of the gain: WGAN variants beat their vanilla-GAN counterparts for every encoder-generator pairing tested.
  • The margin over f-AnoGAN (F1 0.9250 vs 0.9099) indicates that adding an LSTM-plus-attention encoder on top of a WGAN with an inverse-mapping encoder buys a further improvement over the strongest generative baseline.
  • Transformer-only baselines underperform generative ones, supporting the paper's position that for short bursty anomalies, reconstruction-based scoring localizes the event better than long-horizon attention does.

Reading between the lines

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

  • Extension the authors did not run: swapping weight-clipping for the gradient-penalty form of the Wasserstein loss, which the paper itself credits in f-AnoGAN, would probably stretch the reported margin further and is a one-line change worth testing.
  • The reported numbers come from a single run with no variance; a cautious operator should treat the roughly 1.5-point F1 gap over f-AnoGAN as provisional until seed-averaged results rule out run-to-run noise.
  • Because the dataset is proprietary and unreleased, the claim of detecting previously undetected anomalies currently rests on one ISP's traffic; a public backbone trace with labeled packet-loss events would show how far the result travels.
  • The normal-only training recipe should transfer to other silent network faults such as fiber cuts, faulty middleboxes, or routing drops that emit no alarm, since the scoring mechanism, a reconstruction residual in a Wasserstein feature space, does not depend on the failure mode.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 4 minor

Summary. The manuscript proposes WBHT, a Wasserstein GAN augmented with an LSTM-based encoder containing multi-head self-attention and a ConvLSTM generator, for detecting silent black-hole anomalies in backbone network traffic. Training uses an ISP dataset assumed to contain exclusively normal traffic; black-hole labels are used only for test evaluation. The authors report an ablation study in Table I that selects the WGAN/LSTMMultiHead/ConvLSTM configuration, and a comparison in Table II against twelve baselines (including autoencoders, GAN variants, and Transformer-based models), where WBHT achieves the highest F1 score of 0.9250. The central claim is that this architecture outperforms all baseline models for black-hole detection on this dataset.

Significance. The problem is practically important: black-hole anomalies cause silent packet loss in mission-critical networks, and the paper addresses an underexplored application of generative attention models. The use of a real ISP dataset and the breadth of the baseline comparison, including f-AnoGAN, MADGAN, Autoformer, Informer, and TimeSeriesTransformer, are strengths. The ablation in Table I usefully documents the contribution of WGAN versus vanilla GAN and of different encoder/generator choices. If the headline result were reproducible, the architecture would be a modest but useful contribution to semi-supervised anomaly detection. However, the manuscript does not currently provide enough evaluation detail: the anomaly decision rule is unspecified, the loss function in Eq. (3) is not well formed, the selected configuration's metrics in Table I do not match the WBHT row in Table II, and all results appear to be single runs without variance estimates. These gaps prevent independent verification of the central claim.

major comments (5)
  1. [Section III and Section IV-B] Section III states that classification relies on reconstruction errors and D's confidence score, but no anomaly decision rule is given: no threshold, no score normalization, and no statement of whether thresholds were chosen on a validation set, by grid search, or by maximizing test F1. Since every DR/FAR/F1 entry in Table II is a function of this threshold, the headline 'outperforms all baseline models' is not a fixed property of the architecture and cannot be reproduced until the decision rule and threshold-selection procedure are specified.
  2. [Section III, Eq. (3)] Equation 3 is not a well-formed loss function. The first term appears to be (1/n) * sqrt(sum over t of x - G(E(x))) with no square or norm applied to the residual, and the second term similarly lacks a squared norm for the feature residual f(x) - f(G(E(x))). The symbols n and n_d are used in a dimensionally unclear way, and the weighting factor k is not defined beyond 'balancing'. Because the anomaly score is built from reconstruction and feature residuals, the objective must be stated unambiguously, e.g., as a weighted sum of mean-squared-error terms.
  3. [Section IV-A and Section IV-B, Tables I and II] The selected architecture in Table I (WGAN, E: LSTMMultiHead, G: ConvLSTM) is reported with DR 0.9575, FAR 0.0788, F1 0.9261, and Acc 0.9331, but the WBHT row in Table II reports DR 0.9532, FAR 0.0780, F1 0.9250, and Acc 0.9322. The paper does not explain this discrepancy. If these are two runs of the same architecture and data, the protocol should be clarified; if they are different test settings, the model-selection and evaluation procedure needs to be described. In addition, the model configuration is selected as the best F1 row in Table I on the same evaluation setup used for the final comparison, with no held-out validation, so selection bias cannot be ruled out.
  4. [Tables I and II] All results in Tables I and II appear to be single runs without error bars, confidence intervals, or significance tests. The reported F1 improvement over f-AnoGAN is small in relative terms: (0.9250 - 0.9099)/0.9099 is approximately 1.7%. Without multiple seeds or a statistical test, this difference is fragile and may not reflect a real advantage of the proposed architecture. The authors should report mean and standard deviation over several runs and, ideally, provide threshold-robustness curves (e.g., precision-recall or ROC curves) instead of a single operating point.
  5. [Section IV] The paper asserts that the training data contain exclusively normal traffic, with black-hole labels used only for test evaluation, but the dataset is proprietary and not released, and no contamination-verification protocol is described. If the normal-only assumption is violated, the reconstruction-plus-discriminator score would not separate anomalies from normal traffic, undermining the entire evaluation. The authors should state how training windows were verified to be anomaly-free and consider releasing anonymized features or a synthetic benchmark to allow independent checks of this assumption.
minor comments (4)
  1. [Section V] The conclusion contains a typo: 'framewotk' should be 'framework'.
  2. [Section IV-B] The hyperparameter descriptions contain spacing and formatting errors: 'f iltersare', 'numberof attentionheads', 'hiddensize', and 'T − Conv 1D' should be cleaned and made consistent.
  3. [Section III] Reference [22] (Einops) is cited for the multi-head self-attention mechanism, but Einops is a tensor manipulation library, not the source of the MHSA architecture. The authors should cite the original Transformer or attention paper, e.g., Vaswani et al., 2017.
  4. [Section IV-B] The prose mentions 'ConvLSTM-MultiHeadAE' as a baseline, but Table II lists 'ConvMultiHead-AE' and 'LSTMMultiHead-AE'; the naming should be aligned to avoid ambiguity.

Circularity Check

1 steps flagged · score 6.0 of 10

Architecture is selected by maximizing F1 on the labeled test set, so the reported WBHT F1 is the selected maximum and the headline 'outperforms all baselines' is partly forced by the selection procedure.

  1. fitted input called prediction [Section IV and Section IV-A; Table I and Table II]
    "Beyond the previous study, this approach gives a semi-supervised learning performed with all candidate forecasting models and the BH labeled samples were used only in the evaluation of the test set results. ... Among these, the F1 Score was prioritized for model selection ... As a result of these experiments, WBHT was formed ... Overall, WBHT successfully integrates WGAN for stable training, LSTM-based encoding for sequential learning, and Multi-Head Attention for fine-grained feature extraction, allowing it to outperform all baseline models."

    The paper selects the WBHT configuration by maximizing F1 over the Table I grid of GAN-versus-WGAN and E/G architecture combinations, and Table I is evaluated on the same test set whose BH labels are used only for evaluation. The Table II WBHT row (DR 0.9532, FAR 0.0780, F1 0.9250, Acc 0.9322) is exactly the best cell from the WGAN block of Table I, so the reported F1 is the maximum of the candidate configurations on the same labeled test set. Presenting this selected maximum as evidence that WBHT 'outperforms all baseline models' is therefore not an independent prediction of the architecture's merit; it is partly a restatement of the selection criterion.

full rationale

The paper is an empirical benchmark rather than a formal derivation, so most circularity patterns do not apply. The main circular step is the use of the labeled test set for architecture selection: Table I reports F1 for eighteen candidate configurations, F1 is stated as the prioritized selection metric, and the resulting WBHT row is then presented in Table II as the proposed method outperforming all baselines. Because the same test labels drive both selection and evaluation, the headline F1 is partly selected rather than predicted. This is not a self-citation issue: the prior work [4] supplies the dataset collection and cleaning protocol, but the dataset itself comes from an external ISP provider, and the baseline implementations are independent of the authors' prior work, so citing [4] is not load-bearing circularity. The undefined anomaly threshold and the ambiguous loss in Eq. 3 are serious reproducibility and correctness concerns, but they are not circularity: they do not make any claimed result equivalent to its inputs by definition. Overall, the central numerical claim is partially forced by selection-on-test-set leakage, giving a score of 6, while the external baseline comparison and the novelty of the architecture combination still carry some independent content.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central result rests on the normality of the training set, the accuracy of test labels from prior work, and standard WGAN assumptions. It introduces no new physical or mathematical entities, but it does depend on several hand-chosen hyperparameters, including the anomaly threshold and the architecture pair.

free parameters (5)
  • Anomaly decision threshold
    Classification flags sequences that deviate from normal patterns, but the threshold on reconstruction error and discriminator confidence is not specified, so it is a hand-chosen or tuned quantity.
  • Loss weighting factor k in Equation 3
    Balances reconstruction residual and feature residual, but no value or tuning procedure is given.
  • Architecture choice (encoder LSTMMultiHead, generator ConvLSTM) = E: LSTMMultiHead, G: ConvLSTM
    Selected as the best F1 row in Table I; this is model selection on the evaluation setup.
  • Attention heads = 4
    Multi-head attention with 4 heads is mentioned for the LSTMMultiHead-AE baseline and the same module is used in WBHT.
  • Training hyperparameters
    Learning rate, batch size, epochs, latent dimension, and number of LSTM layers are not reported, yet they affect all results.
assumptions (4)
  • domain assumption The training dataset contains exclusively normal network traffic, with BH labels used only for test evaluation.
    The method learns normality from this data; if anomalies contaminate training, the reconstruction-based score loses meaning. Stated in Section IV and Section III.
  • domain assumption Test-set BH labels from the prior data collection [4] are accurate.
    All reported F1 values depend on label quality from the proprietary ISP dataset described in [4].
  • standard math Wasserstein distance with weight clipping provides stable training and a meaningful optimization landscape.
    Assumed from WGAN literature [21]; used as the rationale for choosing WGAN over GAN.
  • ad hoc to paper Reconstruction error plus discriminator feature residual and discriminator confidence jointly separate anomalies from normal traffic.
    The paper states classification relies on reconstruction error and D's confidence score, but no formal justification or threshold procedure is given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of WBHT: A Generative Attention Architecture for Detecting Black Hole Anomalies in Backbone Networks." pith.science (2026). https://pith.science/paper/FEESLUMW

@misc{pith2026250720373,
  author       = {Pith},
  title        = {Pith review of: WBHT: A Generative Attention Architecture for Detecting Black Hole Anomalies in Backbone Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FEESLUMW}},
  note         = {Machine review of arXiv:2507.20373}
}
read the original abstract

We propose the Wasserstein Black Hole Transformer (WBHT) framework for detecting black hole (BH) anomalies in communication networks. These anomalies cause packet loss without failure notifications, disrupting connectivity and leading to financial losses. WBHT combines generative modeling, sequential learning, and attention mechanisms to improve BH anomaly detection. It integrates a Wasserstein generative adversarial network with attention mechanisms for stable training and accurate anomaly identification. The model uses long-short-term memory layers to capture long-term dependencies and convolutional layers for local temporal patterns. A latent space encoding mechanism helps distinguish abnormal network behavior. Tested on real-world network data, WBHT outperforms existing models, achieving significant improvements in F1 score (ranging from 1.65% to 58.76%). Its efficiency and ability to detect previously undetected anomalies make it a valuable tool for proactive network monitoring and security, especially in mission-critical networks.

Figures

Figures reproduced from arXiv: 2507.20373 by the authors.

Figure 1
Figure 1. WBHT: Wasserstein Black Hole Transformer Anomaly Detection Framework. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Comparative Performance Evaluation of BH Detection Models [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 23 canonical work pages

  1. [1]

    Emerging trends in uavs: From placement, semantic communications to generative ai for mission-critical networks,

    Z. Kaleem, F. A. Orakzai, W. Ishaq, K. Latif, J. Zhao, and A. Jamalipour, “Emerging trends in uavs: From placement, semantic communications to generative ai for mission-critical networks,” IEEE Transactions on Consumer Electronics, pp. 1–1, 2024

    work page 2024

  2. [2]

    X-cba: Ex- plainability aided catboosted anomal-e for intrusion detection system,

    K. Kaya, E. Ak, S. Bas, B. Canberk, and S. G. Oguducu, “X-cba: Ex- plainability aided catboosted anomal-e for intrusion detection system,” in ICC 2024 - IEEE International Conference on Communications , 2024, pp. 2288–2293

    work page 2024

  3. [3]

    Investigating on black holes in segment routing networks: Identification and detection,

    M. Polverini, A. Cianfrani, M. Listanti, G. Siano, F. G. Lavacca, and C. C. Campanile, “Investigating on black holes in segment routing networks: Identification and detection,” IEEE Transactions on Network and Service Management , vol. 20, no. 1, pp. 14–29, 2023

    work page 2023

  4. [4]

    Black hole prediction in backbone networks: A com- prehensive and type-independent forecasting model,

    K. Kaya, E. Ak, E. Ozaltun, L. Maglaras, T. Q. Duong, B. Canberk, and S. G. Oguducu, “Black hole prediction in backbone networks: A com- prehensive and type-independent forecasting model,” IEEE Transactions on Network and Service Management , pp. 1–1, 2025

    work page 2025

  5. [5]

    Applications of generative adversarial networks in anomaly detection: A systematic literature review,

    M. Sabuhi, M. Zhou, C.-P. Bezemer, and P. Musilek, “Applications of generative adversarial networks in anomaly detection: A systematic literature review,” IEEE Access, vol. 9, pp. 161 003–161 029, 2021

    work page 2021

  6. [6]

    Bgp anomaly detection techniques: A survey,

    B. Al-Musawi, P. Branch, and G. Armitage, “Bgp anomaly detection techniques: A survey,” IEEE Communications Surveys & Tutorials , vol. 19, no. 1, pp. 377–396, 2017

    work page 2017

  7. [7]

    Deep learning for network intrusion: A hierarchical approach to reduce false alarms,

    S. J. Moore, F. Cruciani, C. D. Nugent, S. Zhang, I. Cleland, and S. Sani, “Deep learning for network intrusion: A hierarchical approach to reduce false alarms,” Intelligent Systems with Applications , vol. 18, p. 200215, 2023

    work page 2023

  8. [8]

    Toward developing efficient conv-ae-based intrusion detection system using heterogeneous dataset,

    M. A. Khan and J. Kim, “Toward developing efficient conv-ae-based intrusion detection system using heterogeneous dataset,” Electronics, vol. 9, no. 11, p. 1771, 2020

    work page 2020

Show all 27 references
  1. [9]

    Knacks of a hybrid anomaly detection model using deep auto-encoder driven gated recurrent unit,

    E. Mushtaq, A. Zameer, and R. Nasir, “Knacks of a hybrid anomaly detection model using deep auto-encoder driven gated recurrent unit,” Computer Networks, vol. 226, p. 109681, 2023

    2023

  2. [10]

    Cannolo: An anomaly detection system based on lstm autoencoders for controller area network,

    S. Longari, D. H. Nova Valcarcel, M. Zago, M. Carminati, and S. Zanero, “Cannolo: An anomaly detection system based on lstm autoencoders for controller area network,” IEEE Transactions on Network and Service Management, vol. 18, no. 2, pp. 1913–1924, 2021

    1913

  3. [11]

    Dct-gan: Dilated convolutional transformer-based gan for time series anomaly detection,

    Y . Li, X. Peng, J. Zhang, Z. Li, and M. Wen, “Dct-gan: Dilated convolutional transformer-based gan for time series anomaly detection,” IEEE Transactions on Knowledge and Data Engineering , vol. 35, no. 4, pp. 3632–3644, 2023

    2023

  4. [12]

    Transec-gan: A transformer-enhanced ids for robust detection and privacy in industrial cps,

    J. Liang, Z. Li, and M. Sadiq, “Transec-gan: A transformer-enhanced ids for robust detection and privacy in industrial cps,” in 2025 IEEE Wireless Communications and Networking Conference (WCNC) , 2025, pp. 1–6

    2025

  5. [13]

    Mul- tivariate time series anomaly detection with adversarial transformer architecture in the internet of things,

    F. Zeng, M. Chen, C. Qian, Y . Wang, Y . Zhou, and W. Tang, “Mul- tivariate time series anomaly detection with adversarial transformer architecture in the internet of things,” Future Generation Computer Systems, vol. 144, pp. 244–255, 2023

    2023

  6. [14]

    Transformer or autoencoder? who is the ultimate adversary for attack detectors?

    S. Laudanna, A. Di Sorbo, P. Vinod, C. A. Visaggio, and G. Canfora, “Transformer or autoencoder? who is the ultimate adversary for attack detectors?” Int. J. Inf. Secur. , vol. 24, no. 1, Nov. 2024. [Online]. Available: https://doi.org/10.1007/s10207-024-00934-9

    2024 doi

  7. [15]

    Transformer-based gan-augmented defender for adversarial usb keystroke injection attacks,

    A. K. Chillara, P. Saxena, and R. R. Maiti, “Transformer-based gan-augmented defender for adversarial usb keystroke injection attacks,” ser. ICDCN ’25. New York, NY , USA: Association for Computing Machinery, 2025, p. 94–103. [Online]. Available: https://doi.org/10.1145/370083...

    2025

  8. [16]

    Next-gen metaverse security through intrusion detection enhanced by transformers and gans,

    Y . Djenouri, A. Nabil Belbachir, A. Belhadi, T. Michalak, and G. Srivas- tava, “Next-gen metaverse security through intrusion detection enhanced by transformers and gans,” IEEE Internet of Things Journal , vol. 12, no. 12, pp. 20 640–20 651, 2025

    2025

  9. [17]

    Feature selection for black hole attacks

    M. B. Yasin, Y . M. Khamayseh, and M. AbuJazoh, “Feature selection for black hole attacks.” J. Univers. Comput. Sci. , vol. 22, no. 4, pp. 521–536, 2016

    2016

  10. [18]

    Blackhole attack detection using machine learning approach on manet,

    S. Pandey and V . Singh, “Blackhole attack detection using machine learning approach on manet,” in 2020 International Conference on Electronics and Sustainable Communication Systems (ICESC) . IEEE, 2020, pp. 797–802

    2020

  11. [19]

    Machine learning models to detect the blackhole attack in wireless adhoc network,

    T. Nagalakshmi, A. Gnanasekar, G. Ramkumar, and A. Sabarivani, “Machine learning models to detect the blackhole attack in wireless adhoc network,” Materials Today: Proceedings , vol. 47, pp. 235–239, 2021

    2021

  12. [20]

    Wsn-ds: A dataset for intrusion detection systems in wireless sensor networks,

    I. Almomani, B. Al-Kasasbeh, and M. Al-Akhras, “Wsn-ds: A dataset for intrusion detection systems in wireless sensor networks,” Journal of Sensors, vol. 2016, 2016

    2016

  13. [21]

    Wasserstein generative ad- versarial networks,

    M. Arjovsky, S. Chintala, and L. Bottou, “Wasserstein generative ad- versarial networks,” in International conference on machine learning . PMLR, 2017, pp. 214–223

    2017

  14. [22]

    Einops: Clear and reliable tensor manipulations with einstein-like notation,

    A. Rogozhnikov, “Einops: Clear and reliable tensor manipulations with einstein-like notation,” in International Conference on Learning Representations, 2022

    2022

  15. [23]

    Unsupervised anomaly detection with generative adversarial networks to guide marker discovery,

    T. Schlegl, P. Seeb ¨ock, S. M. Waldstein, U. Schmidt-Erfurth, and G. Langs, “Unsupervised anomaly detection with generative adversarial networks to guide marker discovery,” in Information Processing in Medical Imaging. Cham: Springer International Publishing, 2017, pp. 146–157

    2017

  16. [24]

    Mad-gan: Multivariate anomaly detection for time series data with generative adversarial networks,

    D. Li, D. Chen, B. Jin, L. Shi, J. Goh, and S.-K. Ng, “Mad-gan: Multivariate anomaly detection for time series data with generative adversarial networks,” in International conference on artificial neural networks. Springer, 2019, pp. 703–716

    2019

  17. [25]

    f-anogan: Fast unsupervised anomaly detection with generative adversarial networks,

    T. Schlegl, P. Seeb ¨ock, S. M. Waldstein, G. Langs, and U. Schmidt- Erfurth, “f-anogan: Fast unsupervised anomaly detection with generative adversarial networks,” Medical image analysis, vol. 54, pp. 30–44, 2019

    2019

  18. [26]

    Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting,

    H. Wu, J. Xu, J. Wang, and M. Long, “Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting,” Advances in neural information processing systems, vol. 34, pp. 22 419– 22 430, 2021

    2021

  19. [27]

    Informer: Beyond efficient transformer for long sequence time-series forecasting,

    H. Zhou, S. Zhang, J. Peng, S. Zhang, J. Li, H. Xiong, and W. Zhang, “Informer: Beyond efficient transformer for long sequence time-series forecasting,” in Proceedings of the AAAI conference on artificial intel- ligence, vol. 35, no. 12, 2021, pp. 11 106–11 115

    2021

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.