Pith. sign in

REVIEW 3 major objections 4 minor 42 references

Evaluating Privacy-Utility Tradeoffs in Synthetic Smart Grid Data

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read The choice of generative model determines whether synthetic smart-meter data is useful or private—and no single model wins at both.

desk verdict The utility benchmark is internally self-referential—full-synthetic results train and test on generated data—so the headline 88.2% macro-F1 is not a real-world claim; the privacy analysis is sounder. read the letter →

arxiv 2506.11026 v1 pith:RV2WSGKI submitted 2025-05-20 cs.LG cs.CY

classification cs.LGcs.CY
keywords syntheticdataprivacy-utilitytradeoffsmartgriddynamictime-of-usetariffsgenerativeadversarialnetworksdiffusionmodelsmembershipinferencereconstructionattack
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 tries to establish that when synthetic data is used to decide which households would benefit from dynamic time-of-use electricity tariffs, the architecture of the generative model sets the privacy–utility trade-off. Comparing WGAN, CTGAN, a diffusion model, and Gaussian noise augmentation under semi- and full-synthetic regimes on a large London smart-meter dataset, it finds that diffusion full-synthetic data gives the best classification utility (macro-F1 up to 88.2%, against 67.5% on real data), while CTGAN full-synthetic gives the strongest reconstruction-attack resistance (privacy-risk score 0.16). WGAN performs worst on both axes, and noise augmentation is useful but leaks almost as much as the raw records. A sympathetic reader would take the paper's point to be that generator choice is a first-order design decision for privacy-preserving energy analytics.

What carries the argument

The argument is carried by two measurement constructions rather than a single theorem. The first is the responsiveness score: a z-scored linear combination of six engineered behavioural features—high- and low-tariff usage ratios, peak-hour ratio, weekend shift, load entropy, and low-tariff load factor—weighted by the first principal component and thresholded at the 75th percentile to produce binary labels (Eqs. 1–2). The second is the privacy-risk score PRS, the ratio of reconstruction-error improvement over a noise baseline to the maximum possible improvement from real data; PRS near 1 means the synthetic table reveals a target household feature about as well as the real table, and PRS near 0 means reconstruction is no better than guessing. These two constructions define the axes of the paper's privacy–utility Pareto plot and make generator comparisons quantitative.

What would settle it

Take a dynamic-tariff trial with observed consumption responses, such as the 2013 London tariff trial comparing usage during high-price windows before and after the tariff, and check whether households the paper's PCA threshold labels 'responsive' actually shift more consumption than the rest. If the labelled group does not shift substantially more, the reported macro-F1 gains (88.2% for diffusion) measure fidelity to the proxy label, not real tariff suitability.

Watch

Extended reading notes

Core claim

The central claim, on the paper's own terms, is that the structural design of the generator—more than how much real data is retained—determines both downstream utility and privacy leakage in synthetic smart-grid data. The authors demonstrate this by deriving a binary 'responsive to dynamic tariffs' label from the thresholded first principal component of engineered behavioural features, then benchmarking five classifiers on semi- and full-synthetic versions of four generators, with privacy measured by membership-inference AUC and a reconstruction-based privacy-risk score. Their headline results are that diffusion full-synthesis reaches 88.2% macro-F1, CTGAN full-synthesis reduces reconstruction leakage to PRS 0.16, semi-synthetic mixtures leak nearly as much as the real table, and WGAN full-synthesis is weakest on both utility and privacy.

Load-bearing premise

The load-bearing premise is that the PCA-derived responsiveness score genuinely identifies households that would benefit from a dynamic tariff; the paper itself states these labels may not align with economic responsiveness or actual behavioural change.

Editorial extensions

If this is right

  • Diffusion full-synthetic data can lift macro-F1 for tariff-suitability classification from 67.5% on real data to 88.2%, so utilities could train accurate classifiers without releasing any real household readings.
  • CTGAN full-synthetic data cuts reconstruction leakage to PRS 0.16 while still supporting good classifiers (macro-F1 up to 82.5%), making it the recommended choice when the synthetic dataset itself will be published.
  • Semi-synthetic sets that mix real and synthetic rows leak almost as much as the raw data (PRS near 1), so privacy-preserving pipelines should prefer full-synthetic replacement over augmentation.
  • WGAN full-synthetic performs worst on both utility and privacy (macro-F1 50.7%, PRS 0.62), which implies that architecture choice, not synthesis regime, is the dominant factor in the trade-off.
  • Membership-inference AUC stays near 0.61–0.64 across all generators, so the practical privacy risk in this setting comes from reconstruction rather than membership disclosure.

Reading between the lines

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

  • A natural next step is a multi-objective search over generator architectures and conditioning strategies, rather than selecting among a few fixed models, to trace the full privacy-utility Pareto frontier.
  • The 75th-percentile threshold and the PCA-based label are arbitrary; a sensitivity analysis across thresholds, or validation against measured consumption shifts from the original tariff trial, would show whether the utility ranking survives a change in how 'responsive' is defined.
  • Because semi-synthetic leakage is near complete, a testable extension is adaptive mixing: adding real rows only up to a PRS budget would let practitioners quantify how much real data can be safely included.
  • Since membership-inference attacks barely differentiate generators, future privacy evaluations of tabular synthetic data should weight reconstruction and attribute-inference attacks more heavily.
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

3 major / 4 minor

Summary. The paper benchmarks four synthetic data generation strategies (WGAN, CTGAN, diffusion models, and Gaussian noise augmentation) under semi-synthetic and full-synthetic regimes for smart-meter household data. Utility is measured by macro-F1 on a binary tariff-responsiveness classification task, fidelity by KL/JS divergence and summary statistics, and privacy by membership-inference and reconstruction attacks. The headline claims are that diffusion full-synthetic data achieves the highest utility (macro-F1 up to 88.2%) and CTGAN full-synthetic offers the strongest reconstruction resistance (PRS=0.16), leading to the conclusion that generator architecture is a first-order determinant of the privacy-utility trade-off.

Significance. If the evaluation were sound, the paper would provide a useful comparative framework for privacy-preserving synthetic smart-grid data, with several concrete strengths: it covers four generators and two synthesis regimes, uses paired significance tests (nominally), reports multiple fidelity and privacy metrics, and the authors state that code and data are public. The main significance, however, is contingent on the validity of the utility measurement. The reported utility numbers rest on two interdependent assumptions: that the PCA-derived responsiveness label (Eqs. 1-2) is a meaningful proxy for tariff suitability, and that full-synthetic utility can be measured by training and testing on synthetic folds. The second assumption is directly contradicted by the stated full-synthetic protocol in Section 4.1 and Algorithm 1. The stress-test concern therefore lands, and it is load-bearing for the paper's central claim.

major comments (3)
  1. [Section 4.1 / Section 4.2 / Algorithm 1 / Table 2] The full-synthetic utility numbers are self-consistency scores, not predictions on real households. Section 4.1 defines full-synthetic as replacing real data entirely with synthetic samples, and Algorithm 1 then splits the current dataset (X,y) into traini and testi. For full-synthetic regimes, both the training and test folds are synthetic, so the reported macro-F1 (e.g., Diffusion Full-Synthetic, 88.2%, Table 2) measures how well the classifier reproduces the generator's internal feature-label mapping. The claim that full-synthetic diffusion 'beats the real baseline by over 20 points' is therefore unsupported: the real baseline is evaluated on real held-out data, while the synthetic condition is evaluated on synthetic held-out data. The paper never states that all regimes are evaluated on a single fixed real test set. This also invalidates the paired significance tests in Section 4.2: the tests are only paired if the test folds are identical across conditions, which they are not under the current protocol. The utility comparison and all conclusions derived from it (RQ2, Section 5.2, Figure 4, Section 6) need to be re-run on a fixed real test set that is held out from both generator training and classifier training.
  2. [Section 3.2 / Section 7] The label construction is circular with respect to the utility claim. Equations (1)-(2) define the binary responsiveness label as the thresholded first principal component of the same engineered behavioural features that are fed to the classifiers. High macro-F1 therefore partly reflects how well a model recovers a known deterministic linear function of its own inputs, not necessarily how well it predicts economically meaningful tariff responsiveness. The authors acknowledge in Section 7 that the labels 'may not align perfectly with economic responsiveness or actual behavioural change' and that no external validation exists. This is not merely a caveat: all utility numbers and the privacy-utility Pareto analysis are measured against this label. The authors should provide external validation (e.g., using the actual dToU trial enrollment or measured demand-shift outcomes as labels) or, failing that, a sensitivity analysis over the threshold q, the feature set, and the PCA loadings; without such evidence, the headline utility results cannot be interpreted as evidence about real-world tariff suitability.
  3. [Section 4.2 / Table 2] The significance-testing protocol is not described rigorously enough to support the arrows in Table 2. The text says 'paired tests across the outer-fold macro-F1 vectors,' but as noted above, the test folds differ between the real, semi-synthetic, and full-synthetic conditions. Additionally, Algorithm 1 does not state that the outer split is stratified, despite Section 3.3.1 claiming an outer 5-fold stratified CV; this matters for a class ratio of 3:1. The authors should clarify the exact fold structure and confirm that the same real test folds are used for every condition, or else revise the significance claims.
minor comments (4)
  1. [Table 2] The XGBoost entry for Diffusion Semi-Synth. appears truncated as '72.0 ± 4.'; several entries in Table 5 contain typographical spacing issues ('0 999', '0 997').
  2. [Section 4.2] The sentence 'We draw niter = 10 random configurations per classifier 5 are evaluated in each outer CV fold' is garbled and should be rewritten.
  3. [Table 1] The 'Best Classifier' column is not defined: it appears to report the best macro-F1 across the five classifiers, but Table 2 already gives the full breakdown and the two tables are not fully consistent (e.g., Table 1 lists XGB for Real while Table 2 reports XGB=67.5, matching, but this convention should be stated explicitly).
  4. [Section 3.4] The WGAN bullet begins 'Wasserstein–GP GAN Our WGAN comprises...' with a missing separator, and there are scattered typographical issues such as 'R ELU' and 'XGB OOST' in the text.

Circularity Check

3 steps flagged · score 7.0 of 10

Full-synthetic utility claims reduce to self-consistency scores: classifiers are trained and tested on synthetic folds produced by the same generator, and diffusion labels come from the generator's own classification head.

  1. fitted input called prediction [Section 4.2 (Algorithm 1) and Section 4.1 (Experimental Protocol)]
    "The semi-synthetic approach augments the original dataset with synthetic observations, while the full-synthetic approach replaces real data entirely with synthetic samples. ... Algorithm 1: ... Split (X, y) into traini, testi ... Evaluate on testi; store macro-F1i"

    For full-synthetic regimes, the (X, y) passed to Algorithm 1 are entirely synthetic, so the held-out test fold is also synthetic and is produced by the same generator that produced the training fold. Macro-F1 then measures how well the classifier reproduces the generator's internal feature-label mapping, not how well it predicts real households. The paper calls this an 'unbiased estimate of downstream utility' and compares the resulting scores against a real-data baseline, but the test distributions are different by construction, so the comparison is not a paired or like-for-like utility measure. The headline Diffusion Full-Synthetic macro-F1 of 88.2% is therefore a self-consistency score, not evidence of real-world predictive utility.

  2. self definitional [Section 3.4, Diffusion Model paragraph]
    "The architecture jointly predicts two outputs: (i) the additive Gaussian noise εθ(xt, t) and (ii) the binary class logit ŷ, using a dual-head design. ... Generated features are decoded through the learned denoising path, and class labels are derived from the class logit head using a thresholded sigmoid output."

    In the full-synthetic regime, the labels that downstream classifiers are trained and tested on are generated by the same diffusion model's own class-logit head. Thus a high macro-F1 indicates that an SVM or XGBoost can recover the generator's internal labeling rule from the generated features, not that the generator produces data with any external responsiveness signal. The target variable is defined by the very model being evaluated, making the claimed 'utility' self-referential and inflated relative to any real-world tariff-suitability construct.

1 more flagged steps
  1. self definitional [Section 3.2, Eqs. (1)-(2) and Section 5.4 (RQ1)]
    "Since explicit ground truth labels for responsiveness are unavailable, we adopt an unsupervised scoring method. ... We then apply Principal Component Analysis (PCA) and extract the first principal component (PC1) ... Responsivei = I[si > Quantileq(s)] ... Our results confirm that SVM and XGBoost achieve the highest classification performance, demonstrating strong predictive capability with a macro-F1 score of 0.88."

    The binary label is constructed as the thresholded first principal component of the same behavioral feature vectors that are later fed to the classifiers. High macro-F1 therefore partly reflects how well a model recovers a known linear function of its own inputs, rather than how well it predicts independently measured economic responsiveness. The paper itself concedes in Section 7 that the labels 'may not align perfectly with economic responsiveness or actual behavioural change,' which confirms the self-referential nature of the evaluation. While the relative ranking of generators might persist under a better label, the absolute utility numbers are not validated against any external criterion.

full rationale

The central utility claim of the paper is that Diffusion Full-Synthetic data achieves the highest macro-F1 (88.2%), outperforming the real-data baseline by over 20 points. On inspection, this number is not a prediction on real households: Section 4.1 defines full-synthetic as replacing real data entirely with synthetic samples, and Algorithm 1 then splits (X, y) into train and test folds. For full-synthetic regimes, both folds are generated by the same generator, so the macro-F1 measures the learnability of the generator's synthetic feature-label distribution, not downstream utility on real data. This is compounded for diffusion, where labels are produced by the model's own class-logit head, making the target definitionally dependent on the generator being evaluated. A separate but related issue is the PCA-derived responsiveness label (Eqs. 1-2): because the label is a thresholded linear combination of the same features used as classifier inputs, part of the measured accuracy is tautological. The paper's own Threats to Validity section acknowledges the label-validity limitation but does not address the full-synthetic test-set circularity. No significant self-citation load-bearing is present; the issue is the evaluation protocol itself. The score is 7 because the headline utility result reduces, by construction, to a self-consistency measure, while the relative ranking of generators and the privacy results retain some independent content.

Assumptions & free parameters 4 free parameters · 3 assumptions · 1 invented entities

The central claim rests on a constructed label and several unspecified experimental details. The PC1 loadings and threshold q are fitted to the evaluation data; the responsiveness score is an invented entity with no independent evidence. Standard machine learning assumptions about cross-validation and KDE are not listed.

free parameters (4)
  • PC1 loadings w_j = estimated from real-data PCA (not reported numerically)
    These weights define the responsiveness score in Eq. (1); all classification labels and utility results depend on them. They are fitted to the same data used for evaluation.
  • Label threshold q = 0.75 (75th percentile of PC1 scores)
    Set by hand in Eq. (2); determines the class split (top 25% responsive). No sensitivity analysis is performed.
  • Gaussian noise scale fraction = unspecified
    In Eq. (5), sigma_x is set as a fraction of each feature's empirical standard deviation; the fraction is not reported, making the noise baseline non-reproducible.
  • WGAN entropy and class-balance regularizer weights = unspecified
    Described qualitatively in Section 3.4(1); no values are given, affecting reproducibility of the WGAN variant.
assumptions (3)
  • domain assumption The first principal component of standardized behavioural features captures tariff responsiveness.
    Section 3.2 states this proxy without external validation; the authors explicitly acknowledge in Section 7 that labels may not align with real responsiveness.
  • domain assumption Responsiveness to dToU tariffs can be inferred from consumption behaviour features alone.
    All features are derived from smart meter readings; no socioeconomic, survey, or intervention outcome data are used.
  • ad hoc to paper Macro-F1 on the PCA-derived label is a meaningful utility measure for tariff suitability.
    The choice of utility metric is contingent on the constructed label being valid; the paper itself lists this as a threat to validity.
invented entities (1)
  • PCA-derived responsiveness score s_i
    purpose: Serves as ground truth label for the classification task (Eq. 1-2).
    No external validation against real tariff response outcomes; the paper acknowledges in Section 7 that this score may not reflect true responsiveness.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Evaluating Privacy-Utility Tradeoffs in Synthetic Smart Grid Data." pith.science (2026). https://pith.science/paper/RV2WSGKI

@misc{pith2026250611026,
  author       = {Pith},
  title        = {Pith review of: Evaluating Privacy-Utility Tradeoffs in Synthetic Smart Grid Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RV2WSGKI}},
  note         = {Machine review of arXiv:2506.11026}
}
read the original abstract

The widespread adoption of dynamic Time-of-Use (dToU) electricity tariffs requires accurately identifying households that would benefit from such pricing structures. However, the use of real consumption data poses serious privacy concerns, motivating the adoption of synthetic alternatives. In this study, we conduct a comparative evaluation of four synthetic data generation methods, Wasserstein-GP Generative Adversarial Networks (WGAN), Conditional Tabular GAN (CTGAN), Diffusion Models, and Gaussian noise augmentation, under different synthetic regimes. We assess classification utility, distribution fidelity, and privacy leakage. Our results show that architectural design plays a key role: diffusion models achieve the highest utility (macro-F1 up to 88.2%), while CTGAN provide the strongest resistance to reconstruction attacks. These findings highlight the potential of structured generative models for developing privacy-preserving, data-driven energy systems.

Figures

Figures reproduced from arXiv: 2506.11026 by the authors.

Figure 1
Figure 1. Proposed methodology overview. 3.1 Data Preprocessing The dataset originates from the UK Power Networks’ Low Carbon London project (2011–2014), comprising 167 million half-hourly electricity consumption records from 5,567 households in Greater London. It includes two groups: approximately 1,100 households en￾rolled in a dToU tariff trial in 2013, receiving real-time price signals, and around 4,500 households on a fi… view at source ↗
Figure 2
Figure 2. Consumption autocorrelation heatmap for a household subsample up to lag 20. Warmer regions indicate stronger correlations, reflecting patterns in energy usage over time. 3.3.1 Baseline Classifiers and Evaluation Protocol We benchmark five standard tabular classifiers: Decision Tree (DT), Random Forest (RF), k-Nearest Neighbours (KNN), Support-Vector Machine with an RBF kernel (SVM), and Extreme Gradient Boosting (XG… view at source ↗
Figure 3
Figure 3. t-SNE (perplexity 30) of real vs. synthetic samples. Colours denote generator + synthesis regime [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: presents the Privacy–Utility Pareto frontier, mapping each generative strategy along two axes: privacy, quantified by the Privacy Risk Score (PRS, x-axis, lower is better), and utility, measured by macro-F1 classification performance (y-axis, higher is better). The fig…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

42 extracted references · 21 canonical work pages

  1. [1]

    Abadi, A

    M. Abadi, A. Chu, I. Goodfellow, H. B. McMahan, I. Mironov, K. Tal- war, and L. Zhang. Deep learning with differential privacy. In Pro- ceedings of the 2016 ACM SIGSAC Conference on Computer and Com- munications Security, CCS ’16, page 308–318, New York, NY , USA,

  2. [2]

    Albert and R

    A. Albert and R. Rajagopal. Smart meter driven segmentation: What your consumption says about you. IEEE Transactions on Power Sys- tems, 28(4):4019–4030, 2013. doi: 10.1109/TPWRS.2013.2266122

  3. [3]

    Alshantti, A

    A. Alshantti, A. Rasheed, and F. Westad. Privacy re-identification at- tacks on tabular gans, 2024. URL https://arxiv.org/abs/2404.00696

  4. [4]

    Beckel, L

    C. Beckel, L. Sadamori, T. Staake, and S. Santini. Revealing household characteristics from smart meter data. Energy, 78:397– 410, 2014. ISSN 0360-5442. doi: https://doi.org/10.1016/j.energy. 2014.10.025. URL https://www.sciencedirect.com/science/article/pii/ S0360544214011748

  5. [5]

    Faraday: Synthetic Smart Meter Generator for the smart grid

    S. Chai and G. Chadney. Faraday: Synthetic smart meter generator for the smart grid, 2024. URL https://arxiv.org/abs/2404.04314

  6. [6]

    Faruqui and S

    A. Faruqui and S. Sergici. Household response to dynamic pricing of electricity: a survey of 15 experiments. Journal of Regulatory Eco- nomics, 38(2):193–225, 2010

  7. [7]

    Freier and V

    J. Freier and V . von Loessl. Dynamic electricity tariffs: Design- ing reasonable pricing schemes for private households. Energy Eco- nomics, 112:106146, 2022. ISSN 0140-9883. doi: https://doi.org/ 10.1016/j.eneco.2022.106146. URL https://www.sciencedirect.com/ science/article/pii/S0140988322003012

  8. [8]

    Goodfellow, J

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio. Generative adversarial networks. Communications of the ACM, 63(11):139–144, 2020

Show all 42 references
  1. [9]

    Gulrajani, F

    I. Gulrajani, F. Ahmed, M. Arjovsky, V . Dumoulin, and A. Courville. Improved training of wasserstein gans, 2017. URL https://arxiv.org/ abs/1704.00028

  2. [10]

    Guo and M

    B. Guo and M. Weeks. Dynamic tariffs, demand response, and regulation in retail electricity markets. Energy Economics , 106: 105774, 2022. ISSN 0140-9883. doi: https://doi.org/10.1016/j.eneco. 2021.105774. URL https://www.sciencedirect.com/science/article/pii/ S0140988321006149

  3. [11]

    R. e. a. Gupta. Tide: Time-series diffusion for forecasting and imputa- tion. arXiv preprint arXiv:2305.13395, 2023

  4. [12]

    J. Ho, A. Jain, and P. Abbeel. Denoising diffusion probabilistic models,

  5. [13]

    Hyeong, J

    J. Hyeong, J. Kim, N. Park, and S. Jajodia. An empirical study on the membership inference attack against tabular data synthesis models,

  6. [14]

    Khorramshahi, H

    P. Khorramshahi, H. Souri, R. Chellappa, and S. Feizi. Gans with variational entropy regularizers: Applications in mitigating the mode- collapse issue, 2020. URL https://arxiv.org/abs/2009.11921

  7. [15]

    D. P. Kingma and J. Ba. Adam: A method for stochastic optimization,

  8. [16]

    Kullback and R

    S. Kullback and R. A. Leibler. On information and sufficiency. Annals of Mathematical Statistics, 22(1):79–86, 1951. URL https://www.jstor. org/stable/2236703

  9. [17]

    Liang, Z

    X. Liang, Z. Wang, and H. Wang. Synthetic data generation for res- idential load patterns via recurrent gan and ensemble method. IEEE Transactions on Instrumentation and Measurement, 73:1–12, 2024. doi: 10.1109/TIM.2024.3480225

  10. [19]

    Maalej and C

    A. Maalej and C. Rebai. Sensor data augmentation strategy for load forecasting in smart grid context. In 2021 18th International Multi- Conference on Systems, Signals & Devices (SSD) , pages 979–983,

  11. [20]

    M. L. Menéndez, J. Pardo, L. Pardo, and M. Pardo. The jensen-shannon divergence. Journal of the Franklin Institute, 334(2):307–318, 1997

  12. [21]

    Mhaske, R

    D. Mhaske, R. Satam, S. Londhe, and et al. An efficient electricity theft detection using xgboost. International Journal of Engineering Applied Sciences and Technology, 6(10):282–287, 2022

  13. [22]

    Mirza and S

    M. Mirza and S. Osindero. Conditional generative adversarial nets. CoRR, abs/1411.1784, 2014. URL http://arxiv.org/abs/1411.1784

  14. [23]

    J. Moon, S. Jung, S. Park, and E. Hwang. Conditional tabular gan- based two-stage data generation scheme for short-term load forecasting. IEEE Access, 8:205327–205339, 2020. doi: 10.1109/ACCESS.2020. 3037063

  15. [24]

    Papernot, M

    N. Papernot, M. Abadi, Úlfar Erlingsson, I. Goodfellow, and K. Tal- war. Semi-supervised knowledge transfer for deep learning from private training data, 2017. URL https://arxiv.org/abs/1610.05755

  16. [25]

    Petrlik, P

    I. Petrlik, P. Lezama, C. Rodriguez, and et al. Electricity theft detection using machine learning. International Journal of Advanced Computer Science and Applications, 13(12):420–428, 2022

  17. [26]

    D. W. Scott. Multivariate Density Estimation: Theory, Practice and Visualization. Wiley, 1992

  18. [27]

    Shokri, M

    R. Shokri, M. Stronati, C. Song, and V . Shmatikov. Membership infer- ence attacks against machine learning models. In2017 IEEE symposium on security and privacy (SP), pages 3–18. IEEE, 2017

  19. [28]

    Shokri, M

    R. Shokri, M. Stronati, C. Song, and V . Shmatikov. Membership in- ference attacks against machine learning models, 2017. URL https: //arxiv.org/abs/1610.05820

  20. [29]

    Sohl-Dickstein, E

    J. Sohl-Dickstein, E. A. Weiss, N. Maheswaranathan, and S. Gan- guli. Deep unsupervised learning using nonequilibrium thermodynam- ics, 2015. URL https://arxiv.org/abs/1503.03585

  21. [30]

    The probable error of a mean

    Student. The probable error of a mean. Biometrika, 6(1):1–25, 1908. doi: 10.1093/biomet/6.1.1

  22. [31]

    Tashiro, J

    Y . Tashiro, J. Song, and S. Ermon. Csdi: Conditional score-based diffu- sion models for imputation. NeurIPS, 2021

  23. [32]

    von Loessl

    V . von Loessl. Smart meter-related data privacy concerns and dy- namic electricity tariffs: Evidence from a stated choice experiment. Energy Policy, 180:113645, 2023. ISSN 0301-4215. doi: https:// doi.org/10.1016/j.enpol.2023.113645. URL https://www.sciencedirect. com/science...

  24. [33]

    Wilcoxon

    F. Wilcoxon. Individual comparisons by ranking methods. Biometrics Bulletin, 1(6):80–83, 1945. doi: 10.2307/3001968

  25. [34]

    X. Wu, Y . Pang, T. Liu, and S. Wu. Winning the midst challenge: New membership inference attacks on diffusion models for tabular data syn- thesis. arXiv preprint arXiv:2503.12008, 2025

  26. [35]

    L. Xu, M. Skoularidou, A. Cuesta-Infante, and K. Veeramachaneni. Modeling tabular data using conditional gan. In Advances in Neural Information Processing Systems, 2019

  27. [36]

    L. Xu, M. Skoularidou, A. Cueto, and J. González. Modeling tabular data using conditional gan. In Advances in Neural Information Process- ing Systems 32, pages 7335–7345, 2019

  28. [37]

    Yilmaz and R

    B. Yilmaz and R. Korn. Synthetic demand data generation for indi- vidual electricity consumers : Generative adversarial networks (gans). Energy and AI , 9:100161, 2022. ISSN 2666-5468. doi: https:// doi.org/10.1016/j.egyai.2022.100161. URL https://www.sciencedirect. com/scienc...

  29. [38]

    J. Yoon, D. Jarrett, and M. Schaar. Time-series generative adversarial networks. 12 2019

  30. [2016]

    ISBN 9781450341394

    Association for Computing Machinery. ISBN 9781450341394. doi: 10.1145/2976749.2978318. URL https://doi.org/10.1145/2976749. 2978318

  31. [2017]

    URL https://arxiv.org/abs/1412.6980

  32. [2020]

    URL https://arxiv.org/abs/2006.11239

  33. [2021]

    doi: 10.1109/SSD52085.2021.9429417

  34. [2022]

    URL https://arxiv.org/abs/2208.08114

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.