arxiv: 2604.20293 · v1 · submitted 2026-04-22 · 💻 cs.LG

Recognition: unknown

Synthetic Flight Data Generation Using Generative Models

Karim Aly , Alexei Sharpanskykh

Authors on Pith no claims yet

Pith reviewed 2026-05-10 00:39 UTC · model grok-4.3

classification 💻 cs.LG

keywords synthetic datagenerative modelsflight delay predictionTVAEGaussian Copulaaviation datapredictive utilitytabular data

0 comments

The pith

Generative models produce synthetic flight data that trains delay prediction models to accuracy levels comparable to real records.

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

The paper tests whether two generative models, Tabular Variational Autoencoder and Gaussian Copula, can create realistic synthetic versions of flight records to address data scarcity and privacy limits in aviation. It applies a four-stage evaluation that checks statistical similarity, fidelity, diversity, and predictive utility for downstream tasks such as forecasting delays. Results show that models trained on the synthetic data reach prediction accuracy close to those trained on actual flight data. This matters because real aviation datasets are often restricted and lack enough examples of rare events like delays or diversions.

Core claim

Tabular Variational Autoencoder and Gaussian Copula models are adapted to generate synthetic flight information. Gaussian Copula achieves higher statistical similarity and fidelity but incurs higher computational cost, while Tabular Variational Autoencoder scales efficiently to large datasets. Both produce data that supports flight delay prediction models with accuracy comparable to models trained on real data, as confirmed through the four-stage assessment.

What carries the argument

The four-stage evaluation framework that quantifies statistical similarity, fidelity, diversity, and predictive utility of synthetic flight data generated by Tabular Variational Autoencoder and Gaussian Copula.

If this is right

Synthetic data can replace or supplement confidential real flight records in model development.
Rare events such as delays can be augmented in training sets without violating data restrictions.
Tabular Variational Autoencoder enables practical generation at scales where Gaussian Copula becomes impractical.
Prediction systems for critical aviation events can be trained and validated without direct access to full real datasets.

Where Pith is reading between the lines

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

The same generation process could be applied to other sparse outcomes like cancellations or diversions.
Public release of synthetic flight datasets might accelerate collaborative research while preserving confidentiality.
Further tests on datasets with even lower frequencies of rare events would clarify the limits of the current evaluation.

Load-bearing premise

That matching statistical similarity, fidelity, diversity, and predictive utility in synthetic data guarantees reliable performance when predicting rare aviation events on real data.

What would settle it

A controlled test in which a flight delay classifier trained solely on synthetic data shows substantially lower accuracy on a held-out set of real flight records than an identical classifier trained on real data.

Figures

Figures reproduced from arXiv: 2604.20293 by Alexei Sharpanskykh, Karim Aly.

**Figure 1.** Figure 1: Overview of the analysis framework. A. Data and Preprocessing This study uses the publicly available “TranStats Database for Airline On-Time Performance” [15] from the Bureau of Transportation Statistics (BTS) [16]. The data covers U.S. domestic flights and provides detailed information on flight delays, cancellations, diversions, and their causes. This level of detail makes it a valuable resource for mode… view at source ↗

**Figure 2.** Figure 2: TVAE - PCA analysis of real (blue) vs. synthetic (red) flight information. Replacing datetime features with numerical time duration values led to a partial improvement in capturing data variability. However, Fig. 2b indicates that certain clusters present in the real dataset remain absent in the synthetic data. This observation is further supported by Fig. 3b, which shows that the generated dataset contai… view at source ↗

**Figure 3.** Figure 3: TVAE - Class balance analysis of real (top) vs. synthetic (bottom) departure and arrival delay labels (1 = delayed, 0 = on time). Another significant distinction between “df utc d” and “df utc d 2” lies in the inclusion of the “Taxi In Time (min)” feature. This feature can be calculated as the time difference between “Actual Arrival Time UTC” and “Wheels On Time UTC”. However, while this feature depends on… view at source ↗

**Figure 4.** Figure 4: GC - PCA analysis of real (blue) vs. synthetic (red) flight information. Unlike TVAE, the CG model demonstrated greater robustness to feature selection and data types across both experimental configurations: “df utc ts” in Experiment 4 and “df utc d 2” in Experiment 5. As shown in [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

**Figure 6.** Figure 6: Similarity of marginal distributions for real (blue) vs. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗

**Figure 5.** Figure 5: GC - Class balance analysis of real (top) vs. synthetic (bottom) departure and arrival delay labels (1 = delayed, 0 = on time). The diversity analysis demonstrated that using the “df utc d 2” DataFrame as input for both the TVAE and GC generative models resulted in improved diversity coverage and a class distribution more closely aligned with the real data. Consequently, the subsequent evaluation will focu… view at source ↗

**Figure 7.** Figure 7: Correlation between distance and air time for real [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗

read the original abstract

The increasing adoption of synthetic data in aviation research offers a promising solution to data scarcity and confidentiality challenges. This study investigates the potential of generative models to produce realistic synthetic flight data and evaluates their quality through a comprehensive four-stage assessment framework. The need for synthetic flight data arises from their potential to serve as an alternative to confidential real-world records and to augment rare events in historical datasets. These enhanced datasets can then be used to train machine learning models that predict critical events, such as flight delays, cancellations, diversions, and turnaround times. Two generative models, Tabular Variational Autoencoder (TVAE) and Gaussian Copula (GC), are adapted to generate synthetic flight information and compared based on their ability to preserve statistical similarity, fidelity, diversity, and predictive utility. Results indicate that while GC achieves higher statistical similarity and fidelity, its computational cost hinders its applicability to large datasets. In contrast, TVAE efficiently handles large datasets and enables scalable synthetic data generation. The findings demonstrate that synthetic data can support flight delay prediction models with accuracy comparable to those trained on real data. These results pave the way for leveraging synthetic flight data to enhance predictive modeling in air transportation.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note private letter to a colleague

This applies TVAE and Gaussian Copula to flight data and claims the synthetic versions train delay predictors with comparable accuracy, but it is domain adaptation of known models with limited evidence on rare-event conditionals.

read the letter

This paper takes two established generative models for tabular data, TVAE and Gaussian Copula, and applies them to produce synthetic flight records. It then checks whether those records can train a delay prediction model at roughly the same accuracy as real data. That is the core result they report. The work is useful because aviation datasets are often restricted by privacy rules and because delays and similar events are rare, so augmentation helps. They run a four-stage check covering statistical similarity, fidelity, diversity, and predictive utility, and they note that TVAE scales better to large sets while GC scores higher on the similarity metrics. That comparison is straightforward and practical. The evaluation framework itself is reasonable for this kind of tabular synthetic data task and gives readers a clear way to judge the output. The main soft spot is whether the generated data actually keeps the conditional relationships that matter for delay prediction. Matching overall distributions and pairwise correlations does not automatically preserve the joint probabilities involving specific combinations of airline, weather, time, and other factors, especially when events are imbalanced. The abstract only says accuracy is comparable without numbers, error bars, or details on how the utility stage was tested against real held-out data. If the full paper shows proper cross-distribution checks and no degradation on rare cases, the claim strengthens; otherwise it risks overstating reliability for downstream use. This is aimed at applied researchers in air transportation who need data augmentation for ML. It is not a new technique in generative modeling, so core ML readers will not find much here. It deserves peer review because the problem is real, the pipeline is concrete, and the evaluation is at least structured, even if the utility evidence needs tightening.

Referee Report

2 major / 2 minor

Summary. The paper investigates the use of generative models, TVAE and Gaussian Copula, to create synthetic flight data for addressing data scarcity and confidentiality in aviation research. It proposes a four-stage evaluation framework covering statistical similarity, fidelity, diversity, and predictive utility. The key finding is that synthetic data can train flight delay prediction models with accuracy comparable to real data, with TVAE offering better scalability than GC for large datasets.

Significance. This work could have practical significance in enabling machine learning applications in air transportation by mitigating privacy concerns and allowing augmentation of rare events. The explicit comparison of two generative approaches and their trade-offs in quality versus efficiency provides useful guidance. However, the absence of detailed quantitative results in the provided abstract limits the immediate impact assessment.

major comments (2)

[Abstract] The assertion that 'synthetic data can support flight delay prediction models with accuracy comparable to those trained on real data' lacks any supporting quantitative metrics, error bars, specific accuracy values, baseline comparisons, or information on data splits and handling of rare events. This omission is critical as it prevents evaluation of whether the predictive utility stage truly validates the central claim or if issues like overfitting are present.
[Four-stage evaluation framework] While the framework includes predictive utility, it is unclear if it tests the preservation of conditional distributions P(delay | covariates) rather than just overall accuracy. Statistical similarity and fidelity metrics (marginals, pairwise) do not guarantee that higher-order dependencies relevant to rare delay events are maintained, which could cause the downstream models to underperform on real data despite passing the four stages.

minor comments (2)

[Abstract] The computational cost comparison for GC is stated qualitatively ('hinders its applicability') without numerical benchmarks such as runtime or memory usage on the dataset size.
Ensure that all acronyms (TVAE, GC) are defined at first use and that the data source and preprocessing pipeline are described in sufficient detail for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The comments highlight important aspects of clarity and evaluation rigor that we have addressed through targeted revisions.

read point-by-point responses

Referee: [Abstract] The assertion that 'synthetic data can support flight delay prediction models with accuracy comparable to those trained on real data' lacks any supporting quantitative metrics, error bars, specific accuracy values, baseline comparisons, or information on data splits and handling of rare events. This omission is critical as it prevents evaluation of whether the predictive utility stage truly validates the central claim or if issues like overfitting are present.

Authors: We agree that the abstract would be strengthened by including key quantitative results. The full manuscript reports these details in the predictive utility experiments (including accuracy comparisons between real and synthetic training data, data split procedures, and handling of the dataset). To address the concern directly, we have revised the abstract to incorporate specific accuracy values, baseline comparisons, and a brief note on the evaluation protocol. revision: yes
Referee: [Four-stage evaluation framework] While the framework includes predictive utility, it is unclear if it tests the preservation of conditional distributions P(delay | covariates) rather than just overall accuracy. Statistical similarity and fidelity metrics (marginals, pairwise) do not guarantee that higher-order dependencies relevant to rare delay events are maintained, which could cause the downstream models to underperform on real data despite passing the four stages.

Authors: The predictive utility stage trains delay prediction models exclusively on synthetic data and evaluates them on real held-out data. This cross-evaluation directly measures whether the synthetic data preserves the dependencies required for accurate downstream prediction, including those involving delay events. We acknowledge that marginal and pairwise metrics alone do not fully capture higher-order conditionals; the predictive utility step serves as the primary safeguard against this. In the revision we have added an explicit discussion of this limitation together with supplementary conditional distribution checks for rare delay events to further substantiate the framework. revision: partial

Circularity Check

0 steps flagged

No circularity detected in empirical generative modeling study

full rationale

The paper conducts an empirical comparison of TVAE and Gaussian Copula models for synthetic flight data generation. It applies a four-stage evaluation (statistical similarity, fidelity, diversity, predictive utility) by training delay predictors on synthetic vs. real data and measuring accuracy on held-out real data. No derivation chain, fitted parameters renamed as predictions, self-definitional metrics, or load-bearing self-citations appear in the described methodology or results. The central claim rests on direct experimental outcomes against external real data benchmarks rather than reducing to its own inputs by construction. This is a standard, non-circular empirical ML evaluation.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No new mathematical derivations, free parameters fitted within the paper, axioms, or invented entities are introduced; the work relies entirely on standard implementations of existing generative models and conventional evaluation metrics from the ML literature.

pith-pipeline@v0.9.0 · 5497 in / 1044 out tokens · 38303 ms · 2026-05-10T00:39:32.498089+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

49 extracted references · 16 canonical work pages · 1 internal anchor

[1]

Survey on synthetic data generation, evaluation methods and gans,

A. Figueira and B. Vaz, “Survey on synthetic data generation, evaluation methods and gans,”Mathematics, vol. 10, no. 15, 2022. [Online]. Available: https://www.mdpi.com/2227-7390/10/15/2733

2022
[2]

The synthetic data vault,

N. Patki, R. Wedge, and K. Veeramachaneni, “The synthetic data vault,” in2016 IEEE International Conference on Data Science and Advanced Analytics (DSAA), 2016, pp. 399–410

2016
[3]

Modeling tabular data using conditional gan,

L. Xu, M. Skoularidou, A. Cuesta-Infante, and K. Veeramachaneni, “Modeling tabular data using conditional gan,” inAdvances in Neural Information Processing Systems, H. Wallach, H. Larochelle, A. Beygelzimer, F. d'Alch ´e-Buc, E. Fox, and R. Garnett, Eds., vol. 32. Curran Associates, Inc., 2019. [Online]. Available: https://proceedings.neurips.cc/paper fil...

2019
[4]

Smote: synthetic minority over-sampling technique,

N. V . Chawla, K. W. Bowyer, L. O. Hall, and W. P. Kegelmeyer, “Smote: synthetic minority over-sampling technique,”Journal of artificial intel- ligence research, vol. 16, pp. 321–357, 2002

2002
[5]

Adasyn: Adaptive synthetic sampling approach for imbalanced learning,

H. He, Y . Bai, E. A. Garcia, and S. Li, “Adasyn: Adaptive synthetic sampling approach for imbalanced learning,” in2008 IEEE interna- tional joint conference on neural networks (IEEE world congress on computational intelligence). Ieee, 2008, pp. 1322–1328

2008
[6]

Auto-Encoding Variational Bayes

D. P. Kingma and M. Welling, “Auto-encoding variational bayes,” in International Conference on Learning Representations (ICLR), 2013. [Online]. Available: https://arxiv.org/abs/1312.6114

work page internal anchor Pith review Pith/arXiv arXiv 2013
[7]

Generative adversarial nets,

I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde- Farley, S. Ozair, A. Courville, and Y . Bengio, “Generative adversarial nets,” inAdvances in Neural Information Processing Systems, vol. 27. Curran Associates, Inc., June 2014. [Online]. Available: https://proceedings.neurips.cc/paper files/paper/ 2014/file/5ca3e9b122f61f8f06494c97b1afccf3-Paper.pdf

2014
[8]

arXiv preprint arXiv:1806.03384 (2018)

N. Park, M. Mohammadi, K. Gorde, S. Jajodia, H. Park, and Y . Kim, “Data synthesis based on generative adversarial networks,”arXiv preprint arXiv:1806.03384, 2018

work page arXiv 2018
[9]

Synthesizing Tabular Data using Generative Adversarial Networks

L. Xu and K. Veeramachaneni, “Synthesizing tabular data using gener- ative adversarial networks,”arXiv preprint arXiv:1811.11264, 2018

work page Pith review arXiv 2018
[10]

Gen- erating multi-label discrete patient records using generative adversarial networks,

E. Choi, S. Biswal, B. Malin, J. Duke, W. F. Stewart, and J. Sun, “Gen- erating multi-label discrete patient records using generative adversarial networks,” inMachine learning for healthcare conference. PMLR, 2017, pp. 286–305

2017
[11]

Veegan: Reducing mode collapse in gans using implicit variational learning,

A. Srivastava, L. Valkov, C. Russell, M. U. Gutmann, and C. Sutton, “Veegan: Reducing mode collapse in gans using implicit variational learning,”Advances in neural information processing systems, vol. 30, 2017

2017
[12]

Deep-learning-aided packet routing in aeronautical ad hoc networks relying on real flight data: From single-objective to near-pareto multiobjective optimization,

D. Liu, J. Zhang, J. Cui, S. X. Ng, R. G. Maunder, and L. H. Hanzo, “Deep-learning-aided packet routing in aeronautical ad hoc networks relying on real flight data: From single-objective to near-pareto multiobjective optimization,”IEEE Internet of Things Journal, vol. 9, pp. 4598–4614, 2021. [Online]. Available: https: //api.semanticscholar.org/CorpusID:238673995

2021
[13]

An exploratory assessment of llm’s potential toward flight trajectory reconstruction analysis,

Q. Zhang and J. H. Mott, “An exploratory assessment of llm’s potential toward flight trajectory reconstruction analysis,”ArXiv, vol. abs/2401.06204, 2024. [Online]. Available: https://api.semanticscholar. org/CorpusID:266977542

work page arXiv 2024
[14]

2024 , keywords =

S. Wijnands, A. Sharpanskykh, and K. Aly, “Generation of synthetic aircraft landing trajectories using generative adversarial networks,” 2024. [Online]. Available: https://zenodo.org/doi/10.5281/zenodo.14774664

work page doi:10.5281/zenodo.14774664 2024
[15]

Transtats database for airline on-time performance,

B. of Transportation Statistics, “Transtats database for airline on-time performance,” 2023, accessed: 2025-01-23. [Online]. Available: https: //transtats.bts.gov/Tables.asp?QO VQ=EFD&QO anzr=Nv4yv0r% FDb0-gvzr%FDcr4s14zn0pr%FDQn6n&QO fu146 anzr=b0-gvzr

2023
[16]

Department of Transportation

U.S. Department of Transportation. Bureau of transportation statistics. [Online]. Available: https://www.bts.gov/
[17]

Towards autonomous cybersecurity: An intelligent automl framework for autonomous intrusion detection,

L. Yang and A. Shami, “Towards autonomous cybersecurity: An intelligent automl framework for autonomous intrusion detection,” inAutonomousCyber@CCS, 2023. [Online]. Available: https://api. semanticscholar.org/CorpusID:272423857

2023
[18]

Towards a framework on tabular synthetic data generation: a minimalist approach: theory, use cases, and limitations,

Y . Shen, A. Sudjianto, R. ArunPrakash, A. Bhattacharyya, M. Rao, Y . Wang, J. Vaughan, and N. Zhou, “Towards a framework on tabular synthetic data generation: a minimalist approach: theory, use cases, and limitations,”ArXiv, vol. abs/2411.10982, 2024. [Online]. Available: https://api.semanticscholar.org/CorpusID:274131324

work page arXiv 2024
[19]

The intriguing properties of model explanations,

M. Al-Shedivat, A. Dubey, and E. P. Xing, “The intriguing properties of model explanations,”arXiv preprint arXiv:1801.09808, 2018

work page arXiv 2018
[20]

Robust variational autoencoder for tabular data with beta divergence,

H. Akrami, S. Ayd ¨ore, R. M. Leahy, and A. A. Joshi, “Robust variational autoencoder for tabular data with beta divergence,” ArXiv, vol. abs/2006.08204, 2020. [Online]. Available: https://api. semanticscholar.org/CorpusID:219687586

work page arXiv 2006
[21]

R. B. Nelsen,An introduction to copulas. Springer, 2006

2006
[22]

Binary gaussian copula synthesis: A novel data augmentation technique to advance ml- based clinical decision support systems for early prediction of dialysis among ckd patients,

H. Khosravi, S. Das, A. Al-Mamun, and I. Ahmed, “Binary gaussian copula synthesis: A novel data augmentation technique to advance ml- based clinical decision support systems for early prediction of dialysis among ckd patients,”ArXiv, vol. abs/2403.00965, 2024. [Online]. Available: https://api.semanticscholar.org/CorpusID:268230538

work page arXiv 2024
[23]

Differentially private release of high-dimensional datasets using the gaussian copula,

H. J. Asghar, M. Ding, T. Rakotoarivelo, S. Mrabet, and M. A. Kˆaafar, “Differentially private release of high-dimensional datasets using the gaussian copula,”ArXiv, vol. abs/1902.01499, 2019. [Online]. Available: https://api.semanticscholar.org/CorpusID:59604403

work page arXiv 1902
[24]

Measuring re-identification risk using a synthetic estimator to enable data sharing,

Y . Jiang, L. Mosquera, B. Jiang, L. Kong, and K. E. Emam, “Measuring re-identification risk using a synthetic estimator to enable data sharing,”PLoS ONE, vol. 17, 2022. [Online]. Available: https://api.semanticscholar.org/CorpusID:249748022

2022
[25]

Copulas documentation,

S. D. Vault, “Copulas documentation,” 2025, accessed: 2025-01-27. [Online]. Available: https://sdv.dev/Copulas/index.html

2025
[26]

Principal component analysis,

S. Wold, K. Esbensen, and P. Geladi, “Principal component analysis,” Chemometrics and intelligent laboratory systems, vol. 2, no. 1-3, pp. 37–52, 1987

1987
[27]

Robust bayesian inference for discrete outcomes with the total variation distance,

J. Knoblauch and L. V omfell, “Robust bayesian inference for discrete outcomes with the total variation distance,”ArXiv, vol. abs/2010.13456,

work page arXiv 2010
[28]

Available: https://api.semanticscholar.org/CorpusID: 225066970

[Online]. Available: https://api.semanticscholar.org/CorpusID: 225066970
[29]

Numerically more stable computation of the p- values for the two-sample kolmogorov-smirnov test,

T. Viehmann, “Numerically more stable computation of the p- values for the two-sample kolmogorov-smirnov test,”arXiv preprint arXiv:2102.08037, 2021

work page arXiv 2021
[30]

Correlationsimilarity,

S. Developers, “Correlationsimilarity,” 2023, accessed: 2025- 02-03. [Online]. Available: https://docs.sdv.dev/sdmetrics/metrics/ metrics-glossary/correlationsimilarity

2023
[31]

Contingencysimilarity,

——, “Contingencysimilarity,” 2023, accessed: 2025-02-03. [On- line]. Available: https://docs.sdv.dev/sdmetrics/metrics/metrics-glossary/ contingencysimilarity

2023
[32]

Random forests,

L. Breiman, “Random forests,”Machine learning, vol. 45, pp. 5–32, 2001

2001
[33]

Greedy function approximation: a gradient boosting machine,

J. H. Friedman, “Greedy function approximation: a gradient boosting machine,”Annals of statistics, pp. 1189–1232, 2001

2001
[34]

Nearest neighbor pattern classification,

T. Cover and P. Hart, “Nearest neighbor pattern classification,”IEEE transactions on information theory, vol. 13, no. 1, pp. 21–27, 1967

1967
[35]

Breiman,Classification and regression trees

L. Breiman,Classification and regression trees. Routledge, 2017

2017
[36]

Naive (bayes) at forty: The independence assumption in information retrieval,

D. D. Lewis, “Naive (bayes) at forty: The independence assumption in information retrieval,” inEuropean conference on machine learning. Springer, 1998, pp. 4–15

1998
[37]

The regression analysis of binary sequences,

D. R. Cox, “The regression analysis of binary sequences,”Journal of the Royal Statistical Society Series B: Statistical Methodology, vol. 20, no. 2, pp. 215–232, 1958

1958
[38]

Sgdclassifier - scikit-learn 1.3.0 documentation,

S. learn Developers, “Sgdclassifier - scikit-learn 1.3.0 documentation,” n.d., accessed: 2025-01-31. [Online]. Available: https://scikit-learn.org/ stable/modules/generated/sklearn.linear model.SGDClassifier.html

2025
[39]

A comparative study of sampling methods with cross-validation in the fedhome framework,

A. Ahmadi, S. S. Sharif, and Y . M. Banad, “A comparative study of sampling methods with cross-validation in the fedhome framework,”ArXiv, vol. abs/2406.01950, 2024. [Online]. Available: https://api.semanticscholar.org/CorpusID:270226221

work page arXiv 2024
[40]

A systematic analysis of performance measures for classification tasks,

M. Sokolova and G. Lapalme, “A systematic analysis of performance measures for classification tasks,”Information processing & manage- ment, vol. 45, no. 4, pp. 427–437, 2009

2009
[41]

Advantages of the mean absolute error (mae) over the root mean square error (rmse) in assessing average model performance,

C. J. Willmott and K. Matsuura, “Advantages of the mean absolute error (mae) over the root mean square error (rmse) in assessing average model performance,”Climate research, vol. 30, no. 1, pp. 79–82, 2005

2005
[42]

Quantifying uncertainty in random forests via confidence intervals and hypothesis tests,

L. Mentch and G. Hooker, “Quantifying uncertainty in random forests via confidence intervals and hypothesis tests,”Journal of Machine Learning Research, vol. 17, no. 26, pp. 1–41, 2016

2016
[43]

An r-squared measure of goodness of fit for some common nonlinear regression models,

A. C. Cameron and F. A. Windmeijer, “An r-squared measure of goodness of fit for some common nonlinear regression models,”Journal of econometrics, vol. 77, no. 2, pp. 329–342, 1997

1997
[44]

Discretized bottleneck in vae: Posterior-collapse-free sequence-to- sequence learning,

Y . Zhao, P. Yu, S. Mahapatra, Q. Su, and C. Chen, “Discretized bottleneck in vae: Posterior-collapse-free sequence-to- sequence learning,”ArXiv, vol. abs/2004.10603, 2020. [Online]. Available: https://api.semanticscholar.org/CorpusID:216056569

work page arXiv 2004
[45]

Invaert networks: A data-driven framework for model synthesis and identifiability analysis,

G. G. Tong, C. A. S. Long, and D. E. Schiavazzi, “Invaert networks: A data-driven framework for model synthesis and identifiability analysis,”Computer Methods in Applied Mechanics and Engineering,
[46]

Available: https://api.semanticscholar.org/CorpusID: 261697481

[Online]. Available: https://api.semanticscholar.org/CorpusID: 261697481
[47]

Meta-optimized joint generative and contrastive learning for sequential recommendation,

Y . Hao, P. Zhao, J. Fang, J. Qu, G. Liu, F. Zhuang, V . S. Sheng, and X. Zhou, “Meta-optimized joint generative and contrastive learning for sequential recommendation,”2024 IEEE 40th International Conference on Data Engineering (ICDE), pp. 705–718, 2023. [Online]. Available: https://api.semanticscholar.org/CorpusID:264426259

2024
[48]

https://arxiv

J. He, D. M. Spokoyny, G. Neubig, and T. Berg-Kirkpatrick, “Lagging inference networks and posterior collapse in variational autoencoders,”ArXiv, vol. abs/1901.05534, 2019. [Online]. Available: https://api.semanticscholar.org/CorpusID:58014132

work page arXiv 1901
[49]

Discouraging posterior collapse in hierarchical variational autoencoders using context,

A. Kuzina and J. M. Tomczak, “Discouraging posterior collapse in hierarchical variational autoencoders using context,”arXiv preprint arXiv:2302.09976, 2023

work page arXiv 2023