Pith. sign in

REVIEW 5 major objections 6 minor 36 references

AVATAR: Adversarial Autoencoders with Autoregressive Refinement for Time Series Generation

T0 review · 5 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read AVATAR claims to outperform existing time series generators by coupling adversarial autoencoders with a teacher-forced autoregressive supervisor.

desk verdict A reasonable incremental time-series generation architecture whose central superiority claim currently rests on a table containing an impossible score, so it deserves peer review only with a demand for corrected numbers. read the letter →

arxiv 2501.01649 v1 pith:3IRTBI5N submitted 2025-01-03 cs.LG cs.AI

MohammadReza EskandariNasab , Shah Muhammad Hamdi , Soukaina Filali Boubrahimi This is my paper · ORCID
classification cs.LGcs.AI
keywords timeseriesgenerationadversarialautoencodersautoregressivelearningteacherforcingdistributionlossdataaugmentationgenerativenetworksmultivariate
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

AVATAR is a framework for generating synthetic multivariate time series that aims to satisfy two goals at once: matching the overall distribution of real sequences and matching the one-step-ahead conditional distribution at every time step. It builds on adversarial autoencoders, which align a latent code with a Gaussian prior, and adds a teacher-forced supervisor network so the decoder learns temporal transitions. The paper reports that on three datasets (stock prices, sinusoidal signals, and appliance energy readings) AVATAR's synthetic data is harder for an LSTM classifier to distinguish from real data and more useful for training a forecaster than data from TimeGAN, GAN, AAE, and forced-teaching baselines. If these results hold, AVATAR would be a stronger default choice for time series data augmentation.

What carries the argument

The carrying mechanism is the combined loss $L_{AE} = L_{Rjoint} + L_{Ad} + L_S + L_D$ and the architecture it couples: an autoencoder whose latent space is matched to a Gaussian prior by a discriminator, plus a supervisor network that, under teacher forcing, predicts $X_t^{AE}$ from both $X_{1:t-1}^{AE}$ and $X_{1:t-2}^{AE}$. The new supervised loss $L_S$ forces the decoder-supervisor stack to reproduce temporal transitions rather than only reconstructing inputs, and the distribution loss $L_D$ (mean absolute error between batch mean and standard deviation of latent codes and prior samples) replaces some of the discriminator's burden in aligning the aggregated posterior. A regularized GRU with batch normalization is used inside the encoder, decoder, and supervisor but deliberately not in the discriminator, preserving the adversarial balance.

What would settle it

Train the same LSTM classifier on output from a deliberately broken generator, for instance real sequences with their time steps shuffled to destroy temporal order, and check whether its resemblance score rises near the maximum 0.5; if the classifier cannot flag such obviously non-temporal data, the resemblance scores in Table 1 would not support the superiority claim.

Watch

Extended reading notes

Core claim

The central claim is that combining an adversarial autoencoder with an autoregressive supervisor, trained jointly through a single combined loss, delivers time series generation that is both more realistic and more practically useful than existing baselines. Concretely, Table 1 shows a 46.86% lower resemblance score and a 20.44% lower predictive fidelity error than TimeGAN across three datasets. The paper attributes this to four design elements: a supervised loss that predicts time step $t$ from both $t-1$ and $t-2$ in closed loop, a distribution loss that matches the mean and standard deviation of each latent batch to the Gaussian prior, GRU layers with batch normalization, and a joint training schedule that combines reconstruction, adversarial, supervised, and distribution losses without per-loss weights.

Load-bearing premise

The load-bearing premise is that the LSTM classifier's ability to separate real from synthetic sequences and the LSTM forecaster's error on synthetic training data directly measure generation quality and practical utility.

Editorial extensions

If this is right

  • On the three datasets tested, AVATAR's resemblance score is 46.86% lower than TimeGAN's, meaning a trained LSTM finds AVATAR's synthetic series harder to separate from real ones.
  • AVATAR's predictive fidelity error is 20.44% lower than TimeGAN's, meaning a forecaster trained on AVATAR's synthetic data predicts real future steps more accurately.
  • Ablations in Table 2 indicate every component (joint training, autoregressive learning, distribution loss, regularized GRU) contributes, with removing joint training or autoregressive learning degrading scores the most.
  • AVATAR's stability claim is that it yields consistent results across repeated training sessions, unlike GAN-based baselines.

Reading between the lines

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

  • The distribution loss, which matches batch mean and standard deviation of latent codes to the Gaussian prior, is not specific to time series and could be dropped into any adversarial autoencoder to tighten prior alignment; the paper does not explore this transfer.
  • The two-step supervised loss may reduce exposure bias in the decoder-supervisor stack; testing it on autoregressive forecasting tasks, not just generation, would separate that effect from the adversarial component.
  • The resemblance and predictive-fidelity scores are computed with LSTM probes chosen by the authors; a stronger or different probe family could change the ranking, so the 46.86% and 20.44% figures should be read as conditioned on that protocol.
  • Because the paper frames missing-value imputation as future work, the same joint-training architecture could plausibly be adapted to condition generation on observed subsequences.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper introduces AVATAR, a generative framework for multivariate time series that combines an adversarial autoencoder (AAE) with an autoregressive supervisor, trained jointly via a combined loss. The framework adds a supervised loss over lags t-1 and t-2, a distribution loss that matches the batch mean and standard deviation of latent codes to a Gaussian prior, and a regularized GRU with batch normalization. The authors evaluate AVATAR on three datasets (Energy, Google Stock, and a synthetic sinusoid) against TimeGAN, GAN, AAE, Teacher Forcing, and Professor Forcing, reporting lower resemblance scores and lower predictive fidelity errors. The central claim is that AVATAR consistently surpasses existing time series generation benchmarks.

Significance. If the reported results are correct, AVATAR offers a plausible and useful combination of adversarial autoencoders with autoregressive learning, and the ablation study in Table 2 provides internal evidence that each proposed component contributes to performance. The paper is clearly written and the loss functions are described in detail. However, the central numerical claim is currently undermined by an impossible table entry (P-Forcing Energy resemblance score 5.0±0.0) and by an apparent inconsistency in the joint reconstruction loss equation. The evaluation also lacks statistical significance tests, uses only three datasets (one of which is trivial), and does not validate the classifier-based metric against a broken generator. These issues need to be addressed before the claim of consistent superiority can be accepted.

major comments (5)
  1. [Table 1] Table 1, P-Forcing row: the resemblance score of 5.0±0.0 for the Energy dataset is impossible under the metric defined in §5.1, where the score is computed as 0.5 minus a classification error and is therefore bounded in [-0.5, 0.5]. This implausible entry suggests either a misreported value, a different evaluation protocol, or a bug in the evaluation script; because the headline improvement of 46.86% over TimeGAN is derived from this table, the numerical basis of the central claim is insecure until the entry is corrected and the table recomputed.
  2. [§4.4, Eq. (4.10)] The joint reconstruction loss LRjoint includes the term \sum_t ||Xt - s(X_AE_t)||^2, which is inconsistent with the autoregressive description in §4.1 (where the supervisor predicts X_AE_t from X_AE_{t-1} and X_AE_{t-2}). As written, s takes the current autoencoder output as input, which would not teach temporal transitions; please correct the equation or clarify the actual architecture, as this is a central component of the proposed joint training.
  3. [§5.1 and §5.2] No statistical significance tests are reported, and the standard deviations are extremely small (e.g., 0.001), so the claim that AVATAR 'significantly outperforms' baselines is not supported. Additionally, the resemblance score is never validated against a deliberately broken generator; a weak or miscalibrated LSTM classifier could produce artificially low scores, and the reader cannot tell whether the reported gaps reflect genuine distributional fidelity. Please add significance tests (e.g., paired t-tests or confidence intervals) and a sanity check with a clearly poor generator.
  4. [§5, Benchmark Datasets] The evaluation is limited to three datasets, one of which is a synthetic sinusoid with a trivial deterministic structure. To support the general claim that AVATAR 'consistently surpasses existing benchmarks,' the paper needs more diverse, challenging real-world datasets and a sensitivity analysis of the slicing-window hyperparameter, which is fixed at 24 for all datasets.
  5. [§5, Baseline Methods] SeriesGAN (reference [28], from the same research group) is cited in the related work but is not included in the experimental comparisons. Since SeriesGAN also combines adversarial and autoregressive learning and is closely related to AVATAR, its absence is a notable omission for the claim of consistent superiority over existing methods.
minor comments (6)
  1. [§5, repository link] The URL 'https://github.com/samresume/A V ATAR' contains spaces and is not a valid hyperlink; please provide the correct repository address.
  2. [§5.1] The phrase 'which is then subtracted from 0.5, setting the optimal score to 0 rather than 0.5' is ambiguous; please rewrite it to clarify that the score is 0.5 minus the classification error, so that a score of 0 corresponds to random-chance classification.
  3. [§4.3] The statement that the encoder is trained with twice as many iterations as the discriminator is a training hyperparameter; this is not mentioned in the 'no hyperparameters for loss weighting' claim in §4.4, which could mislead readers.
  4. [Reference [29]] Reference [29] (TimeGAN) gives the DOI/URL https://doi.org/10.1145/3559540, which actually points to a different article (the ACM Computing Surveys systematic review on GANs in time series). Please update the citation to the correct NeurIPS 2019 publication.
  5. [Throughout] The model name appears inconsistently as 'A V ATAR' due to formatting; please use 'AVATAR' consistently.
  6. [§4.2] The justification for excluding statistics such as minimum and maximum because the data is normalized to [0,1] is not logically sound; these quantities remain informative about the shape of the distribution even under normalization. Consider removing this justification or providing a more rigorous rationale.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: AVATAR's gains are measured against external baselines, and its self-citations are non-load-bearing design justifications.

full rationale

The paper's central claim, that AVATAR consistently surpasses existing time series generation benchmarks, is supported by held-out evaluations against external baselines (TimeGAN, GAN, AAE, T-Forcing, P-Forcing). No evaluation metric is incorporated into the training objective: the combined loss in Eq. (4.3), LAE = LRjoint + LAd + LS + LD, consists of reconstruction, adversarial, supervised, and distribution terms, none of which is defined in terms of the resemblance or predictive-fidelity scores of Section 5. The resemblance score is measured post hoc using a separate LSTM classifier, and the predictive fidelity score is measured with an LSTM forecaster trained on synthetic data and evaluated on original data; these are external probes rather than fitted targets. There are self-citations, notably the first author's master's thesis [10] for the general data-augmentation framing and the same group's SeriesGAN [28] to justify choosing GRU over LSTM and RNN in Section 4.3. These are minor and non-load-bearing for the main empirical comparison, which applies identical hyperparameters across all methods and reports results against external benchmarks. The paper therefore does not reduce its predictions to its inputs by construction. Two non-circular concerns should be noted separately: Table 1 reports an impossible resemblance score of 5.0 ± 0.0 for P-Forcing on Energy, which cannot be produced by the metric as defined in Section 5.1 (a classification error subtracted from 0.5 is bounded in [-0.5, 0.5]), and the repository link contains spaces, preventing independent numerical verification. These are reproducibility and data-integrity issues, not circularity.

Assumptions & free parameters 2 free parameters · 7 assumptions · 0 invented entities

The central claim rests on standard GAN latent-space assumptions, the two-lag teacher-forcing heuristic, the mean-std distribution loss heuristic, and the validity of LSTM-based evaluation metrics. No new physical entities are introduced.

free parameters (2)
  • Network hyperparameters = not reported
    Hidden dimensions, number of GRU layers, batch size, learning rate, and iteration counts are said to be identical across baselines but are not listed in the paper; performance likely depends on them.
  • Slicing window size = 24
    Energy and stock series are converted to samples with a window of length 24 (Section 5); this fixes the temporal context the model sees and is not varied.
assumptions (7)
  • standard math Adversarial training with a perfect discriminator minimizes a Jensen-Shannon divergence between real and generated sequence distributions.
    Used to justify the global objective in Section 3, Eq. (3.1); standard GAN theory invoked without proof.
  • domain assumption A zero-mean, unit-variance Gaussian is an appropriate prior for latent codes of all datasets.
    Section 4.2 fixes the prior; no evidence that a different latent prior would not improve generation.
  • ad hoc to paper Supervising the decoder at lags t-1 and t-2 teaches the conditional distribution p(Xt|X1:t-1) without overfitting to immediate next-step dynamics.
    Section 4.1, Eqs. (4.4)-(4.5); the two-lag design is asserted and not compared to one-lag or longer-lag variants.
  • ad hoc to paper Matching batch mean and standard deviation of latent codes to the prior is sufficient to improve distribution alignment.
    Section 4.2, Eqs. (4.6)-(4.8); higher-order statistics are ignored and the distribution-loss ablation shows only small gains.
  • domain assumption LSTM classification error and LSTM predictive fidelity are valid proxies for generation quality and practical utility.
    Sections 5.1-5.2 define the metrics without validating them against human judgment, likelihood, or downstream tasks.
  • domain assumption The three chosen datasets represent the variety of real time series generation challenges.
    Section 5 includes two real datasets and one synthetic sine dataset; no rare-event, medical, or highly non-stationary data are included.
  • ad hoc to paper Batch normalization inside GRU layers helps the encoder, decoder, and supervisor, while excluding it from the discriminator maintains training balance.
    Section 4.3 asserts this design choice without a sensitivity analysis or theoretical justification.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AVATAR: Adversarial Autoencoders with Autoregressive Refinement for Time Series Generation." pith.science (2026). https://pith.science/paper/3IRTBI5N

@misc{pith2026250101649,
  author       = {Pith},
  title        = {Pith review of: AVATAR: Adversarial Autoencoders with Autoregressive Refinement for Time Series Generation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3IRTBI5N}},
  note         = {Machine review of arXiv:2501.01649}
}
read the original abstract

Data augmentation can significantly enhance the performance of machine learning tasks by addressing data scarcity and improving generalization. However, generating time series data presents unique challenges. A model must not only learn a probability distribution that reflects the real data distribution but also capture the conditional distribution at each time step to preserve the inherent temporal dependencies. To address these challenges, we introduce AVATAR, a framework that combines Adversarial Autoencoders (AAE) with Autoregressive Learning to achieve both objectives. Specifically, our technique integrates the autoencoder with a supervisor and introduces a novel supervised loss to assist the decoder in learning the temporal dynamics of time series data. Additionally, we propose another innovative loss function, termed distribution loss, to guide the encoder in more efficiently aligning the aggregated posterior of the autoencoder's latent representation with a prior Gaussian distribution. Furthermore, our framework employs a joint training mechanism to simultaneously train all networks using a combined loss, thereby fulfilling the dual objectives of time series generation. We evaluate our technique across a variety of time series datasets with diverse characteristics. Our experiments demonstrate significant improvements in both the quality and practical utility of the generated data, as assessed by various qualitative and quantitative metrics.

Figures

Figures reproduced from arXiv: 2501.01649 by the authors.

Figure 1
Figure 1. The figure illustrates the AVATAR framework [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. t-SNE visualizations illustrate the distribution [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. PCA visualizations depict the alignment in distribution between the original (red) and syn￾thetic (blue) data samples produced by AVATAR and TimeGAN across three datasets. 0 5 10 15 20 Time 0.375 0.380 0.385 0.390 0.395 0.400 0.405 Amplitude Original Stock 1 0 5 10 15 20 Time 0.140 0.145 0.150 0.155 0.160 0.165 Amplitude Original Stock 2 0 5 10 15 20 Time 0.390 0.395 0.400 0.405 0.410 0.415 0.420 0.425 0.430 Amplitu… view at source ↗
Figures from the paper (2 more)
Figure 6
Figure 6. Figure 6: PCA plots illustrate the degree of alignment 0.380 0.385 Am 0.140 0.145 Am 0390 0.395 0.400 Am 0.075 0.080 Am [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: The figure displays four randomly selected [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 18 canonical work pages

  1. [28]

    M. R. EskandariNasab, S. M. Hamdi, and S. F. Boubrahimi, SeriesGAN: Time Series Gen- eration via Adversarial and Autoregressive Learning , arXiv preprint, arXiv:2410.21203, 2024. [Online]. Available: https://arxiv.org/abs/2410.21203

    work page Pith review arXiv 2024

  2. [1]

    Ahmadzadeh, B

    A. Ahmadzadeh, B. Aydin, M. K. Georgoulis, D. J. Kempton, S. S. Mahajan, and R. A. Angryk, How to Train Your Flare Prediction Model: Revisiting Robust Sampling of Rare Events , The Astrophysical Journal Supplement Series, vol. 254, no. 2, pp. 23, May 2021. [Online]. Available: http://dx.doi.org/10.3847/1538- 4365/abec88

    doi:10.3847/1538- 2021

  3. [2]

    EskandariNasab, S

    M. EskandariNasab, S. M. Hamdi, and S. F. Boubrahimi, Impacts of data preprocessing Copyright © 2025 by SIAM Unauthorized reproduction of this article is prohibited and sampling techniques on solar flare prediction from multivariate time series data of photospheric magnetic field parameters , Astrophysical Journal Supplement Series, in press. [Online]. Av...

    work page doi:10.3847/1538-4365/ad7c4a 2025

  4. [3]

    M. R. EskandariNasab, S. M. Hamdi, and S. F. Boubrahimi, Enhancing Multivariate Time Series-based Solar Flare Prediction with Multifaceted Preprocessing and Contrastive Learning , arXiv preprint, arXiv:2409.14016, 2024. [Online]. Available: https://arxiv.org/abs/2409.14016

    work page Pith review arXiv 2024

  5. [4]

    EXCON: Extreme Instance-based Contrastive Representation Learning of Severely Imbalanced Multivariate Time Series for Solar Flare Prediction

    O. Vural, S. M. Hamdi, and S. F. Boubrahimi, EXCON: Extreme Instance-based Contrastive Repre- sentation Learning of Severely Imbalanced Multivari- ate Time Series for Solar Flare Prediction , arXiv preprint, arXiv:2411.11249, 2024. [Online]. Available: https://arxiv.org/abs/2411.11249

    work page Pith review arXiv 2024

  6. [5]

    Abouelmehdi, A

    K. Abouelmehdi, A. Beni-Hssane, H. Khaloufi, and M. Saadi, Big data security and privacy in healthcare: A Review , Procedia Computer Sci- ence, vol. 113, pp. 73–80, 2017. [Online]. Available: https://doi.org/10.1016/j.procs.2017.08.292

    work page doi:10.1016/j.procs.2017.08.292 2017

  7. [6]

    C. J. Cascalheira, K. Corro, C. Hong, T. K. Rohleen, O. Trac, and M. Beikzadeh, An Analysis of Mpox Communication on Reddit vs Twitter Dur- ing the 2022 Mpox Outbreak , Sexuality Research and Social Policy, Dec. 2024. [Online]. Available: https://doi.org/10.1007/s13178-024-01058-4

    work page doi:10.1007/s13178-024-01058-4 2022

  8. [7]

    Chapagain, Y

    S. Chapagain, Y. Zhao, T. K. Rohleen, S. M. Hamdi, S. F. Boubrahimi, and R. E. Flinn, Predictive Insights into LGBTQ+ Minority Stress: A Transductive Explo- ration of Social Media Discourse , 2024 IEEE 11th In- ternational Conference on Data Science and Advanced Analytics (DSAA), 2024, pp. 1–9. [Online]. Available: https://doi.org/10.1109/DSAA61799.2024.10722807

    arXiv 2024

Show all 36 references
  1. [8]

    Vincent, S

    E. Vincent, S. Watanabe, A. A. Nugraha, J. Barker, and R. Marxer, An analysis of environment, mi- crophone and data simulation mismatches in ro- bust speech recognition , Computer Speech & Lan- guage, vol. 46, pp. 535–557, 2017. [Online]. Available: https://doi.org/10.1016/j.c...

    2017 doi

  2. [9]

    Miko lajczyk and M

    A. Miko lajczyk and M. Grochowski, Data aug- mentation for improving deep learning in im- age classification problem , in Proc. 2018 In- ternational Interdisciplinary PhD Workshop (IIPhDW), 2018, pp. 117–122. [Online]. Available: https://doi.org/10.1109/IIPHDW.2018.8388338

    2018

  3. [10]

    M. R. EskandariNasab, Supervised Generative Adversarial Networks for Time Series Gener- ation in Embedding Space , Master’s Thesis, Utah State University, 2024. [Online]. Available: https://digitalcommons.usu.edu/etd2023/387

    2024

  4. [11]

    Voulodimos, N

    A. Voulodimos, N. Doulamis, A. Doulamis, and E. Pro- topapadakis, Deep Learning for Computer Vision: A Brief Review , Computational Intelligence and Neuro- science, vol. 2018, no. 1, pp. 7068349, 2018. [Online]. Available: https://doi.org/10.1155/2018/7068349

    2018 doi

  5. [12]

    Mehrad, E

    A. Mehrad, E. Nourani, and A. Golzari Oskouei, ADA-COVID: Adversarial Deep Domain Adaptation- Based Diagnosis of COVID-19 from Lung CT Scans Using Triplet Embeddings , Computational Intelligence and Neuroscience, vol. 2022, Article ID 2564022, 17 pages, 2022. [Online]. Availab...

    2022 doi

  6. [13]

    P. Li, P. Hosseinzadeh, O. Bahri, S. F. Boubrahimi, and S. M. Hamdi, Adversarial Attack Driven Data Augmentation for Time Series Classi- fication, in Proc. 2023 International Confer- ence on Machine Learning and Applications (ICMLA), 2023, pp. 653–658. [Online]. Available: htt...

    2023

  7. [14]

    Bahri, P

    O. Bahri, P. Li, P. Hosseinzadeh, S. F. Boubrahimi, and S. M. Hamdi, Denoising Optimization-Based Counterfactual Explanations for Time Series Clas- sification, Pattern Recognition, Springer Nature Switzerland, 2025, pp. 162–179. [Online]. Available: https://doi.org/10.1007/978...

    2025 doi

  8. [15]

    Alshammari, S

    K. Alshammari, S. M. Hamdi, and S. F. Boubrahimi, Transformer Model for Multivariate Time Series Classification: A Case Study of Solar Flare Pre- diction, Pattern Recognition, Springer Nature Switzerland, 2025, pp. 238–254. [Online]. Available: https://doi.org/10.1007/978-3-03...

    2025 doi

  9. [16]

    EskandariNasab, Z

    M. EskandariNasab, Z. Raeisi, R. A. Lashaki, and H. Najafi, A GRU–CNN model for auditory attention detection using microstate and recurrence quantifica- tion analysis , Scientific Reports, 14 (2024), p. 8861, doi: 10.1038/s41598-024-58886-y

    2024 doi

  10. [17]

    Bengio, O

    S. Bengio, O. Vinyals, N. Jaitly, and N. Shazeer, Sched- uled sampling for sequence prediction with recurrent neural networks, in Proc. 28th Int. Conf. Neural Infor- mation Processing Systems, 1 (2015), pp. 1171–1179

    2015

  11. [19]

    Bahdanau, P

    D. Bahdanau, P. Brakel, K. Xu, A. Goyal, R. Lowe, J. Pineau, A. Courville, and Y. Bengio, An actor-critic algorithm for sequence prediction , arXiv preprint arXiv:1607.07086, 2017. [Online]. Available: https://arxiv.org/abs/1607.07086

    2017 arXiv

  12. [20]

    I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Ben- gio, Generative adversarial networks , arXiv preprint arXiv:1406.2661, 2014

    2014 arXiv

  13. [21]

    Mogren, C-RNN-GAN: Continuous recur- rent neural networks with adversarial training , CoRR, abs/1611.09904, 2016

    O. Mogren, C-RNN-GAN: Continuous recur- rent neural networks with adversarial training , CoRR, abs/1611.09904, 2016. [Online]. Available: http://arxiv.org/abs/1611.09904

    2016 arXiv

  14. [22]

    Esteban, S

    C. Esteban, S. L. Hyland, and G. R¨ atsch, Real- valued (Medical) time series generation with recurrent conditional GANs , arXiv e-prints, Jun. 2017, doi: 10.48550/arXiv.1706.02633

    2017 doi arXiv

  15. [23]

    Makhzani, J

    A. Makhzani, J. Shlens, N. Jaitly, I. Goodfel- low, and B. Frey, Adversarial Autoencoders , arXiv:1511.05644, 2016. [Online]. Available: Copyright © 2025 by SIAM Unauthorized reproduction of this article is prohibited https://arxiv.org/abs/1511.05644

    2016 arXiv

  16. [24]

    R. J. Williams and D. Zipser, A learning algo- rithm for continually running fully recurrent neu- ral networks , Neural Computation, vol. 1, no. 2, pp. 270–280, Jun. 1989. [Online]. Available: https://doi.org/10.1162/neco.1989.1.2.270

    1989 doi

  17. [25]

    Ganin, E

    Y. Ganin, E. Ustinova, H. Ajakan, P. Germain, H. Larochelle, F. Laviolette, M. Marchand, and V. Lempitsky, Domain-adversarial training of neural networks, Journal of Machine Learning Research, vol. 17, no. 1, pp. 2096–2030, 2016

    2016

  18. [26]

    A. Lamb, A. Goyal, Y. Zhang, S. Zhang, A. Courville, and Y. Bengio, Professor forcing: A new algo- rithm for training recurrent networks , arXiv preprint arXiv:1610.09038 [stat.ML], 2016. [Online]. Available: https://arxiv.org/abs/1610.09038

    2016 arXiv

  19. [29]

    J. Yoon, D. Jarrett, and M. van der Schaar, Time- series generative adversarial networks , in Advances in Neural Information Processing Systems, 2019. [Online]. Available: https://doi.org/10.1145/3559540

    2019 doi

  20. [30]

    D. P. Kingma and M. Welling, Auto-Encoding Vari- ational Bayes , arXiv preprint arXiv:1312.6114, 2022. [Online]. Available: https://arxiv.org/abs/1312.6114

    2022 arXiv

  21. [31]

    Creswell and A

    A. Creswell and A. A. Bharath, Denoising Ad- versarial Autoencoders , IEEE Transactions on Neu- ral Networks and Learning Systems, vol. 30, no. 4, pp. 968–984, 2019. [Online]. Available: https://doi.org/10.1109/TNNLS.2018.2852738

    2019

  22. [32]

    M. L. Men´ endez, J. A. Pardo, L. Pardo, and M. C. Pardo, The Jensen-Shannon divergence , Journal of the Franklin Institute, vol. 334, no. 2, pp. 307–318, 1997. [Online]. Available: https://doi.org/10.1016/S0016-0032(96)00063-4

    1997 doi

  23. [33]

    Shlens, Notes on Kullback-Leibler Divergence and Likelihood, arXiv preprint arXiv:1404.2000, 2014

    J. Shlens, Notes on Kullback-Leibler Divergence and Likelihood, arXiv preprint arXiv:1404.2000, 2014. [On- line]. Available: https://arxiv.org/abs/1404.2000

    2000 arXiv

  24. [34]

    L. M. Ibarra Candanedo, V. Feldheim, and D. Dera- maix, Data driven prediction models of energy use of appliances in a low-energy house , Energy and Build- ings, vol. 140, pp. 81–97, 2017. [Online]. Available: https://doi.org/10.1016/j.enbuild.2017.01.083

    2017 doi

  25. [35]

    Theis, A

    L. Theis, A. van den Oord, and M. Bethge, A note on the evaluation of generative models , arXiv preprint arXiv:1511.01844, 2016. [Online]. Available: https://arxiv.org/abs/1511.01844

    2016 arXiv

  26. [36]

    Y. Wu, Y. Burda, R. Salakhutdinov, and R. Grosse, On the Quantitative Analysis of Decoder-Based Gen- erative Models, arXiv preprint arXiv:1611.04273, 2017. [Online]. Available: https://arxiv.org/abs/1611.04273

    2017 arXiv

  27. [37]

    van der Maaten and G

    L. van der Maaten and G. Hinton, Visualizing data using t-SNE , Journal of Machine Learning Re- search, 9 (2008), pp. 2579–2605. [Online]. Available: http://jmlr.org/papers/v9/vandermaaten08a.html

    2008

  28. [38]

    F. B. Bryant and P. R. Yarnold, Principal-components analysis and exploratory and confirmatory factor anal- ysis, in Reading and Understanding Multivariate Statistics, L. G. Grimm and P. R. Yarnold, Eds., Amer- ican Psychological Association, Washington, DC, 1995, pp. 99–136. ...

    1995

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.