pith. sign in

arxiv: 2606.26053 · v1 · pith:3APNEQ36new · submitted 2026-06-24 · 📊 stat.ML · cs.LG

When Does Synthetic Data Augmentation Improve Score-Based Imbalanced Classification?

Zhengchi Ma , Pengfei Lyu , Anru R. Zhang This is my paper

Pith reviewed 2026-06-25 19:01 UTC · model grok-4.3

classification 📊 stat.ML cs.LG
keywords synthetic data augmentationclass imbalancescore-based classificationAUROCAUPRCwell-specified modelsminimax boundsmisspecification
0
0 comments X

The pith

Under well-specified score models, synthetic augmentation cannot fundamentally improve population-level performance on imbalance metrics beyond possible variance reduction.

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

The paper builds a framework to characterize when synthetic minority augmentation improves metrics such as AUROC, AUPRC, best-threshold balanced accuracy, and F1. It decomposes the augmentation effect into a shift in effective class weighting and a mismatch between synthetic and true minority distributions. In well-specified score models the raw estimator already recovers the likelihood-ratio ordering that is optimal for these metrics at the population level, so augmentation adds no fundamental gain and can only reduce finite-sample variance while risking extra bias. Minimax lower bounds confirm the raw estimator attains the optimal regret rate in this regime. Under misspecification, however, the weighting change can correct ranking errors that the raw imbalanced objective induces.

Core claim

Under well-specified score models, the raw estimator already targets the likelihood-ratio ordering, which is population-optimal for the metrics considered. Consequently, augmentation cannot provide a fundamental population-level improvement beyond possible finite-sample variance reduction, and may introduce additional bias through synthetic distributional error. We further establish minimax lower bounds showing that the raw estimator already achieves the optimal metric-regret rate in the well-specified regime. Under misspecification, however, augmentation can play a qualitatively different role: by changing the effective class balance, it can alter the restricted-class projection and correct

What carries the argument

Decomposition of augmentation effects into effective class weighting change and synthetic-true minority distributional discrepancy, used to derive improvement conditions and bounds.

Load-bearing premise

The effect of augmentation can be separated into a change in effective class weighting and a discrepancy between the synthetic and true minority distributions.

What would settle it

A controlled simulation with a known well-specified score model where the augmented estimator's metric value is compared to the raw estimator after accounting for finite-sample variance; consistent outperformance by the augmented version would falsify the optimality claim.

Figures

Figures reproduced from arXiv: 2606.26053 by Anru R. Zhang, Pengfei Lyu, Zhengchi Ma.

Figure 1
Figure 1. Figure 1: Flowchart of the main results. best-threshold balanced accuracy, the raw estimator already achieves the optimal rate up to dimension-dependent factors. Third, we identify a qualitatively different phenomenon under model misspecification. When the model class is too restrictive, augmentation is not guaranteed to help, because generating more data cannot remove intrinsic approximation error. In contrast, whe… view at source ↗
Figure 2
Figure 2. Figure 2: AUROC, AUPRC, validation-tuned best-threshold balanced accuracy and F [PITH_FULL_IMAGE:figures/full_fig_p028_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: AUROC and AUPRC as functions of the synthetic proportion, for logistic re [PITH_FULL_IMAGE:figures/full_fig_p030_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: complements the ranking-metric results by comparing validation-tuned best￾threshold balanced accuracy and F1 score. For the validation-tuned best-threshold metrics, adding synthetic data provides little to no improvement across the four data-generating distributions. Performance at the fixed 0.5 threshold is also reported for comparison. For the fixed threshold, synthetic augmentation can substantially imp… view at source ↗
Figure 5
Figure 5. Figure 5: AUROC (top) and AUPRC (bottom) as a function of the synthetic proportion, for [PITH_FULL_IMAGE:figures/full_fig_p031_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Balanced accuracy (top) and F1 score (bottom) at a fixed 0.5 threshold and at a validation-tuned threshold, as a function of the synthetic proportion, for an MLP applied to the t5 distribution (left) and the Gaussian AR(1) distribution with ρ = 0.5 (right) [PITH_FULL_IMAGE:figures/full_fig_p032_6.png] view at source ↗
read the original abstract

Synthetic data augmentation is widely used to mitigate class imbalance, but its theoretical effects on score-based classification remain poorly understood. This paper develops a framework for characterizing when synthetic minority augmentation can improve threshold-integrated and threshold-optimized metrics, including AUROC, AUPRC, best-threshold balanced accuracy, and best-threshold \(\F_1\) score. We separate the effect of augmentation into two components: a change in effective class weighting and a discrepancy between the synthetic and true minority distributions. Under well-specified score models, the raw estimator already targets the likelihood-ratio ordering, which is population-optimal for the metrics considered. Consequently, augmentation cannot provide a fundamental population-level improvement beyond possible finite-sample variance reduction, and may introduce additional bias through synthetic distributional error. We further establish minimax lower bounds showing that the raw estimator already achieves the optimal metric-regret rate in the well-specified regime. Under misspecification, however, augmentation can play a qualitatively different role: by changing the effective class balance, it can alter the restricted-class projection and correct ranking errors induced by the raw imbalanced objective. We provide explicit improvement bounds quantifying the roles of approximation error, finite-sample estimation error, and synthetic distributional error. Simulation studies corroborate the theory, demonstrating limited gains under well-specification and nontrivial but nonmonotone improvements under misspecification.

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 develops a framework for analyzing synthetic minority augmentation in score-based imbalanced classification. It decomposes augmentation effects into effective class weighting and synthetic-true minority distributional discrepancy. Under well-specified score models, the raw estimator targets the likelihood-ratio ordering (population-optimal for AUROC, AUPRC, best-threshold balanced accuracy, and F1), so augmentation yields no fundamental population-level gain beyond possible finite-sample variance reduction and may add bias; minimax lower bounds are established showing the raw estimator attains the optimal metric-regret rate. Under misspecification, augmentation can improve by altering the restricted-class projection and correcting ranking errors, with explicit bounds on approximation, estimation, and distributional errors. Simulations support the claims.

Significance. If the derivations hold, the work supplies rigorous conditions distinguishing when augmentation helps versus harms for these metrics, with practical implications for imbalanced learning. The population-level optimality argument, explicit improvement bounds, and minimax lower bounds (if verified) constitute a clear strength, offering falsifiable predictions grounded in standard MLE consistency rather than ad-hoc assumptions.

major comments (2)
  1. [Abstract and framework section] The decomposition into effective class weighting and distributional discrepancy (abstract) is load-bearing for both the well-specified optimality claim and the misspecification improvement bounds. If residual interaction terms between weighting and discrepancy remain after the separation, the explicit improvement bounds and the conclusion that augmentation cannot improve the ordering under well-specification do not follow.
  2. [Minimax analysis section] The minimax lower-bound claim (abstract) that the raw estimator achieves the optimal metric-regret rate relies on the population maximizer recovering the true likelihood ratio. The specific rate derivation and its dependence on the metric definitions (e.g., threshold-integrated vs. threshold-optimized) must be checked to ensure the bound is not an artifact of the well-specified parametric assumption.
minor comments (2)
  1. Clarify notation for the score model and the exact form of the metrics (AUROC, AUPRC, etc.) at first use to avoid ambiguity in the population vs. empirical distinction.
  2. [Simulation studies] The simulation section should report the precise degree of misspecification used and confirm that the observed nonmonotone improvements align quantitatively with the derived bounds.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments. Below we respond point-by-point to the major comments.

read point-by-point responses

  1. Referee: [Abstract and framework section] The decomposition into effective class weighting and distributional discrepancy (abstract) is load-bearing for both the well-specified optimality claim and the misspecification improvement bounds. If residual interaction terms between weighting and discrepancy remain after the separation, the explicit improvement bounds and the conclusion that augmentation cannot improve the ordering under well-specification do not follow.

    Authors: The decomposition in Section 3 is exact: the population limit of the augmented estimator separates additively into the effective weighting shift and the distributional discrepancy term, with all cross terms vanishing under the score-model normalization. This separation is used to establish both the well-specified optimality (no ordering improvement possible) and the misspecification improvement bounds. We will add one clarifying sentence in the framework section noting that the interaction terms are identically zero under the maintained assumptions. revision: partial

  2. Referee: [Minimax analysis section] The minimax lower-bound claim (abstract) that the raw estimator achieves the optimal metric-regret rate relies on the population maximizer recovering the true likelihood ratio. The specific rate derivation and its dependence on the metric definitions (e.g., threshold-integrated vs. threshold-optimized) must be checked to ensure the bound is not an artifact of the well-specified parametric assumption.

    Authors: The minimax lower bounds are derived under the well-specified parametric regime in which the MLE recovers the true likelihood ratio (the population maximizer for all four metrics). Separate regret rates are obtained for threshold-integrated metrics (AUROC, AUPRC) and threshold-optimized metrics (best-threshold balanced accuracy, F1), and the resulting lower bounds match the upper bounds attained by the raw estimator, confirming tightness rather than an artifact of the parametric assumption. revision: no

Circularity Check

0 steps flagged

No significant circularity; claims follow from standard MLE consistency under well-specification

full rationale

The paper's derivation separates augmentation into effective weighting and synthetic discrepancy components, then invokes the standard result that under correct score-model specification the MLE recovers the true likelihood-ratio ordering (population-optimal for the listed metrics). This is an external property of maximum-likelihood estimation, not a quantity fitted from the same data used to evaluate the metrics or defined in terms of the augmentation effect itself. The minimax lower-bound claim is presented as an information-theoretic consequence of that optimality, and the misspecification case is handled by explicit approximation-error bounds; none of these steps reduce by construction to the paper's own inputs or to self-citations whose content is unverified.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on the domain assumption that score models can be meaningfully classified as well-specified or misspecified and that augmentation effects admit a clean two-component decomposition; no free parameters or invented entities are introduced in the abstract.

axioms (2)
  • domain assumption Score models are well-specified in the primary analysis regime
    This assumption underpins the claim that the raw estimator already achieves population-optimal likelihood-ratio ordering.
  • domain assumption Augmentation effects separate into effective class weighting change and synthetic-true distribution discrepancy
    This decomposition is invoked to derive when augmentation improves metrics versus when it cannot.

pith-pipeline@v0.9.1-grok · 5767 in / 1544 out tokens · 34880 ms · 2026-06-25T19:01:04.319866+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

152 extracted references · 26 canonical work pages · 8 internal anchors

  1. [1]

    Journal of Big Data , volume=

    Impact of random oversampling and random undersampling on the performance of prediction models developed using observational health data , author=. Journal of Big Data , volume=. 2024 , publisher=

    2024

  2. [2]

    BMC bioinformatics , volume=

    SMOTE for high-dimensional class-imbalanced data , author=. BMC bioinformatics , volume=. 2013 , publisher=

    2013

  3. [3]

    Proceedings of the Seventeenth International Joint Conference on Artificial Intelligence , year=

    The Foundations of Cost-Sensitive Learning , author=. Proceedings of the Seventeenth International Joint Conference on Artificial Intelligence , year=

  4. [4]

    Optimal Transport for Applied Mathematicians: Calculus of Variations, PDEs, and Modeling , author=

  5. [5]

    2018 , publisher=

    High-dimensional probability: An introduction with applications in data science , author=. 2018 , publisher=

    2018

  6. [6]

    2019 , publisher=

    High-dimensional statistics: A non-asymptotic viewpoint , author=. 2019 , publisher=

    2019

  7. [7]

    Journal of Chemical Information and Modeling , volume=

    GHOST: adjusting the decision threshold to handle imbalanced data in machine learning , author=. Journal of Chemical Information and Modeling , volume=. 2021 , publisher=

    2021

  8. [8]

    IEEE Access , volume=

    Thors: An efficient approach for making classifiers cost-sensitive , author=. IEEE Access , volume=. 2019 , publisher=

    2019

  9. [9]

    Journal of the American Statistical Association , volume=

    Least ambiguous set-valued classifiers with bounded error levels , author=. Journal of the American Statistical Association , volume=. 2019 , publisher=

    2019

  10. [10]

    The Annals of Statistics , pages=

    Ranking and Empirical Minimization of U-Statistics , author=. The Annals of Statistics , pages=. 2008 , publisher=

    2008

  11. [11]

    The Stata Journal , volume=

    When to consult precision-recall curves , author=. The Stata Journal , volume=. 2020 , publisher=

    2020

  12. [12]

    The Annals of Applied Statistics , year=

    Boosting data analytics with synthetic volume expansion , author=. The Annals of Applied Statistics , year=

  13. [13]

    Nature , volume=

    AI models collapse when trained on recursively generated data , author=. Nature , volume=. 2024 , publisher=

    2024

  14. [14]

    arXiv preprint arXiv:2304.10283 , year=

    Is augmentation effective to improve prediction in imbalanced text datasets? , author=. arXiv preprint arXiv:2304.10283 , year=

    work page arXiv

  15. [15]

    Advances in neural information processing systems , volume=

    Modeling tabular data using conditional gan , author=. Advances in neural information processing systems , volume=

  16. [16]

    Statistica Sinica , volume=

    One Step to Efficient Synthetic Data , author=. Statistica Sinica , volume=

  17. [17]

    arXiv preprint arXiv:2503.21968 , year=

    GLM Inference with AI-Generated Synthetic Data Using Misspecified Linear Regression , author=. arXiv preprint arXiv:2503.21968 , year=

    work page arXiv

  18. [18]

    International Journal of Knowledge Engineering and Soft Data Paradigms , volume=

    Borderline over-sampling for imbalanced data classification , author=. International Journal of Knowledge Engineering and Soft Data Paradigms , volume=. 2011 , publisher=

    2011

  19. [19]

    Information Sciences , volume=

    Geometric SMOTE a geometric-based synthetic oversampling method for imbalanced classification , author=. Information Sciences , volume=. 2019 , publisher=

    2019

  20. [20]

    arXiv preprint arXiv:1711.00837 , year=

    Oversampling for imbalanced learning based on K-means and SMOTE , author=. arXiv preprint arXiv:1711.00837 , year=

    work page arXiv

  21. [21]

    European conference on principles of data mining and knowledge discovery , pages=

    SMOTEBoost: Improving prediction of the minority class in boosting , author=. European conference on principles of data mining and knowledge discovery , pages=. 2003 , organization=

    2003

  22. [22]

    ACM SIGKDD explorations newsletter , volume=

    A study of the behavior of several methods for balancing machine learning training data , author=. ACM SIGKDD explorations newsletter , volume=. 2004 , publisher=

    2004

  23. [23]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    A road to classification in high dimensional space: the regularized optimal affine discriminant , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 2012 , publisher=

    2012

  24. [24]

    Advances in neural information processing systems , volume=

    On spectral clustering: Analysis and an algorithm , author=. Advances in neural information processing systems , volume=

  25. [25]

    IEEE journal of biomedical and health informatics , volume=

    Synthetic patient data generation and evaluation in disease prediction using small and imbalanced datasets , author=. IEEE journal of biomedical and health informatics , volume=. 2022 , publisher=

    2022

  26. [26]

    Nature Medicine , volume=

    Self-improving generative foundation model for synthetic medical image generation and clinical applications , author=. Nature Medicine , volume=. 2025 , publisher=

    2025

  27. [27]

    Nature Medicine , volume=

    Generative models improve fairness of medical classifiers under distribution shifts , author=. Nature Medicine , volume=. 2024 , publisher=

    2024

  28. [28]

    Journal of artificial intelligence research , volume=

    SMOTE for learning from imbalanced data: progress and challenges, marking the 15-year anniversary , author=. Journal of artificial intelligence research , volume=

  29. [29]

    arXiv preprint arXiv:2510.26046 , year=

    Bias-Corrected Data Synthesis for Imbalanced Learning , author=. arXiv preprint arXiv:2510.26046 , year=

    work page arXiv

  30. [30]

    , author=

    Estimating causal effects of treatments in randomized and nonrandomized studies. , author=. Journal of Educational Psychology , volume=. 1974 , publisher=

    1974

  31. [31]

    Statistics in Medicine , volume=

    Stratification and weighting via the propensity score in estimation of causal treatment effects: A comparative study , author=. Statistics in Medicine , volume=. 2004 , publisher=

    2004

  32. [32]

    The Annals of Statistics , volume=

    Semi-supervised inference: General theory and estimation of means , author=. The Annals of Statistics , volume=

  33. [33]

    2008 , organization=

    He, Haibo and Bai, Yang and Garcia, Edwardo A and Li, Shutao , booktitle=. 2008 , organization=

    2008

  34. [34]

    arXiv preprint arXiv:2502.11323 , year=

    A statistical theory of overfitting for imbalanced classification , author=. arXiv preprint arXiv:2502.11323 , year=

    work page arXiv

  35. [35]

    Chawla, Nitesh V and Bowyer, Kevin W and Hall, Lawrence O and Kegelmeyer, W Philip , journal=

  36. [36]

    A theoretical distribution analysis of synthetic minority oversampling technique (

    Elreedy, Dina and Atiya, Amir F and Kamalov, Firuz , journal=. A theoretical distribution analysis of synthetic minority oversampling technique (. 2024 , publisher=

    2024

  37. [37]

    Do we need rebalancing strategies?

    Sakho, Abdoulaye and Malherbe, Emmanuel and Scornet, Erwan , journal=. Do we need rebalancing strategies?

  38. [38]

    Asymptotic behavior of

    Kamalov, Firuz , journal=. Asymptotic behavior of

  39. [39]

    New England Journal of Medicine , volume=

    Machine learning in medicine , author=. New England Journal of Medicine , volume=. 2019 , publisher=

    2019

  40. [40]

    Orphanet Journal of Rare Diseases , volume=

    Diagnosis support systems for rare diseases: a scoping review , author=. Orphanet Journal of Rare Diseases , volume=. 2020 , publisher=

    2020

  41. [41]

    2018 2nd International Conference on Data Science and Business Analytics (ICDSBA) , pages=

    Effect of class imbalanceness in detecting automobile insurance fraud , author=. 2018 2nd International Conference on Data Science and Business Analytics (ICDSBA) , pages=. 2018 , organization=

    2018

  42. [42]

    International Conference on Parallel Problem Solving from Nature , pages=

    Improving imbalanced classification by anomaly detection , author=. International Conference on Parallel Problem Solving from Nature , pages=. 2020 , organization=

    2020

  43. [43]

    Internet of Things , volume=

    A machine learning based robust prediction model for real-life mobile phone data , author=. Internet of Things , volume=. 2019 , publisher=

    2019

  44. [44]

    2020 11th International Conference on Information and Communication Systems (ICICS) , pages=

    Machine learning with oversampling and undersampling techniques: Overview study and experimental results , author=. 2020 11th International Conference on Information and Communication Systems (ICICS) , pages=. 2020 , organization=

    2020

  45. [45]

    A generalization of the k-

    Tomek, Ivan , journal=. A generalization of the k-. 1976 , publisher=

    1976

  46. [46]

    Journal of Machine Learning Research , volume=

    Imbalanced-learn: A python toolbox to tackle the curse of imbalanced datasets in machine learning , author=. Journal of Machine Learning Research , volume=

  47. [47]

    1994 , publisher=

    An introduction to the bootstrap , author=. 1994 , publisher=

    1994

  48. [48]

    Borderline-

    Han, Hui and Wang, Wen Yuan and Mao, Bing Huan , booktitle=. Borderline-. 2005 , organization=

    2005

  49. [49]

    Safe-level-

    Bunkhumpornpat, Chumphol and Sinapiromsaran, Krung and Lursinsap, Chidchanok , booktitle=. Safe-level-. 2009 , organization=

    2009

  50. [50]

    arXiv preprint arXiv:2504.07426 , year=

    Conditional data synthesis augmentation , author=. arXiv preprint arXiv:2504.07426 , year=

    work page arXiv

  51. [51]

    Zhang, Hongyi and Cisse, Moustapha and Dauphin, Yann N and Lopez-Paz, David , journal=

  52. [52]

    How does

    Zhang, Linjun and Deng, Zhun and Kawaguchi, Kenji and Ghorbani, Amirata and Zou, James , journal=. How does

  53. [53]

    When and how

    Zhang, Linjun and Deng, Zhun and Kawaguchi, Kenji and Zou, James , booktitle=. When and how. 2022 , organization=

    2022

  54. [54]

    Obtaining well calibrated probabilities using

    Naeini, Mahdi Pakdaman and Cooper, Gregory and Hauskrecht, Milos , booktitle=. Obtaining well calibrated probabilities using

  55. [55]

    The Annals of Statistics , pages=

    Bayesian inference for causal effects: The role of randomization , author=. The Annals of Statistics , pages=. 1978 , publisher=

    1978

  56. [56]

    Biometrika , volume=

    The central role of the propensity score in observational studies for causal effects , author=. Biometrika , volume=. 1983 , publisher=

    1983

  57. [57]

    IEEE Transactions on Knowledge and Data Engineering , volume=

    A survey on transfer learning , author=. IEEE Transactions on Knowledge and Data Engineering , volume=. 2009 , publisher=

    2009

  58. [58]

    IEEE Transactions on Knowledge and Data Engineering , volume=

    A survey on multi-task learning , author=. IEEE Transactions on Knowledge and Data Engineering , volume=. 2021 , publisher=

    2021

  59. [59]

    Machine Learning , volume=

    Multitask learning , author=. Machine Learning , volume=. 1997 , publisher=

    1997

  60. [60]

    Proceedings of the 27th ACM International Conference on Information and Knowledge Management , pages=

    Imbalanced sentiment classification with multi-task learning , author=. Proceedings of the 27th ACM International Conference on Information and Knowledge Management , pages=

  61. [61]

    arXiv preprint arXiv:2509.23915 , year=

    Revisit the Imbalance Optimization in Multi-task Learning: An Experimental Analysis , author=. arXiv preprint arXiv:2509.23915 , year=

    work page arXiv

  62. [62]

    Multi-modal multi-task learning for joint prediction of clinical scores in

    Zhang, Daoqiang and Shen, Dinggang , booktitle=. Multi-modal multi-task learning for joint prediction of clinical scores in. 2011 , organization=

    2011

  63. [63]

    Journal of the American Statistical Association , volume=

    Analysis of semiparametric regression models for repeated outcomes in the presence of missing data , author=. Journal of the American Statistical Association , volume=. 1995 , publisher=

    1995

  64. [64]

    Political Analysis , volume=

    An introduction to the augmented inverse propensity weighted estimator , author=. Political Analysis , volume=. 2010 , publisher=

    2010

  65. [65]

    2017 , publisher=

    Classification and regression trees , author=. 2017 , publisher=

    2017

  66. [66]

    2000 , publisher=

    Finite mixture models , author=. 2000 , publisher=

    2000

  67. [67]

    Advances in neural information processing systems , volume=

    Denoising diffusion probabilistic models , author=. Advances in neural information processing systems , volume=

  68. [68]

    Score-Based Generative Modeling through Stochastic Differential Equations

    Score-based generative modeling through stochastic differential equations , author=. arXiv preprint arXiv:2011.13456 , year=

    work page internal anchor Pith review Pith/arXiv arXiv 2011

  69. [69]

    Flow Matching for Generative Modeling

    Flow matching for generative modeling , author=. arXiv preprint arXiv:2210.02747 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  70. [70]

    Advances in neural information processing systems , volume=

    Neural ordinary differential equations , author=. Advances in neural information processing systems , volume=

  71. [71]

    Advances in neural information processing systems , volume=

    Generative adversarial nets , author=. Advances in neural information processing systems , volume=

  72. [72]

    Communications of the ACM , volume=

    Generative adversarial networks , author=. Communications of the ACM , volume=. 2020 , publisher=

    2020

  73. [73]

    International conference on machine learning , pages=

    Variational inference with normalizing flows , author=. International conference on machine learning , pages=. 2015 , organization=

    2015

  74. [74]

    Auto-encoding variational

    Kingma, Diederik P and Welling, Max , journal=. Auto-encoding variational

  75. [75]

    Survey on synthetic data generation, evaluation methods and

    Figueira, Alvaro and Vaz, Bruno , journal=. Survey on synthetic data generation, evaluation methods and. 2022 , publisher=

    2022

  76. [76]

    arXiv preprint arXiv:2302.04062 , year=

    Machine learning for synthetic data generation: a review , author=. arXiv preprint arXiv:2302.04062 , year=

    work page arXiv

  77. [77]

    LeCun, Yann , journal=. The

  78. [78]

    arXiv preprint arXiv:2209.11215 , year=

    Sampling is as easy as learning the score: theory for diffusion models with minimal data assumptions , author=. arXiv preprint arXiv:2209.11215 , year=

    work page arXiv

  79. [79]

    , author=

    The meaning and use of the area under a receiver operating characteristic (ROC) curve. , author=. Radiology , volume=

  80. [80]

    Pattern recognition letters , volume=

    An introduction to ROC analysis , author=. Pattern recognition letters , volume=. 2006 , publisher=

    2006

Showing first 80 references.