arxiv: 2605.02390 · v1 · submitted 2026-05-04 · 📡 eess.SY · cs.SY

Recognition: 2 theorem links

· Lean Theorem

Differentially Private Synthetic Voltage Phasor Release for Distribution Grids

Andrew Campbell , Chenyue Zhang , Anna Scaglione , Eli Kerr , Merilyn Chesler , Sean Peisert

Authors on Pith no claims yet

Pith reviewed 2026-05-08 19:00 UTC · model grok-4.3

classification 📡 eess.SY cs.SY

keywords differentially private synthetic datavoltage phasorsdistribution gridsAC power flowadmittance matrixgrid foundation modelsload trajectories

0 comments

The pith

Releasing voltage phasors from differentially private synthetic loads protects the privacy of the grid admittance matrix without any added noise.

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

The paper shows how to generate synthetic voltage phasor trajectories for distribution grids that satisfy differential privacy. It begins by training a differentially private generative model on historical customer load data. Synthetic load trajectories from this model are then fed into the AC power flow equations using the true bus admittance matrix to produce voltages. The randomness already present in the private loads proves sufficient to extend differential privacy guarantees to the voltage outputs with respect to the admittance matrix. This keeps the underlying power flow statistics intact, unlike approaches that perturb the matrix or add noise directly to voltages, and supports training of grid machine learning models.

Core claim

The authors establish that voltage phasors generated by passing differentially private synthetic load trajectories through the exact AC power flow equations on the true admittance matrix satisfy (ε, δ)-differential privacy guarantees with respect to the admittance matrix. Privacy for the network parameters follows directly from the existing randomness in the DP loads, with the bound determined by the chosen adjacency definition, the Jacobian of the AC power flow, and the covariance of the synthetic loads. The resulting trajectories preserve the statistical properties of AC power flow, enabling their use for applications such as Grid Foundation Model training without the distortions that come

What carries the argument

The propagation of DP synthetic loads through the deterministic AC power flow equations, where the Jacobian and load covariance control the sensitivity of voltages to changes in the admittance matrix.

If this is right

A corpus of voltage trajectories can be released that preserves AC power flow statistics for downstream machine learning tasks.
Network topology privacy is obtained as a byproduct of load privacy with no extra mechanism required.
The method outperforms Gaussian perturbation of voltage outputs in empirical tests for training utility.

Where Pith is reading between the lines

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

The same input-level privacy might automatically protect model parameters in other simulation-driven data releases where a deterministic physical map connects inputs to outputs.
Extensions could test whether the bound remains useful when the adjacency definition is relaxed to cover more realistic neighboring grid configurations.
Utilities might share larger synthetic datasets for AI training if this approach generalizes beyond voltage phasors.

Load-bearing premise

The adjacency definition for neighboring admittance matrices, together with the Jacobian and covariance values, allows the privacy bound to transfer from loads to voltages without any separate noise step.

What would settle it

Voltage phasor distributions produced from two adjacent admittance matrices under the same DP load model that can be distinguished with probability exceeding the (ε, δ) bound given the Jacobian and covariance.

Figures

Figures reproduced from arXiv: 2605.02390 by Andrew Campbell, Anna Scaglione, Chenyue Zhang, Eli Kerr, Merilyn Chesler, Sean Peisert.

**Figure 1.** Figure 1: Two DP Paradigms for GFMs. Top is the proposed DP-GMM view at source ↗

**Figure 2.** Figure 2: Wasserstein-1 metric as a function of view at source ↗

**Figure 3.** Figure 3: Training (left) and test (right) MSE on the masked entries of the per-bus missing-data recovery task (Section V-C), plotted against training epoch. All view at source ↗

read the original abstract

Training machine learning models, including Grid Foundation Models (GFMs), requires large volumes of realistic grid data, yet substantial privacy concerns discourage utilities and data providers from sharing load profiles and network parameters. We study the release of synthetic voltage phasor trajectories for distribution grids under differential privacy (DP). We first fit a DP generative model to historical customer loads, then propagate synthetic load trajectories through the AC power flow equations on the true admittance matrix to produce voltage phasors. The central question is whether the randomness already present in the DP synthetic loads is sufficient to protect not only the loads, but also the network topology encoded by the bus admittance matrix. We show that it is. The implication is that a corpus of voltage trajectories can be constructed from DP synthetic loads while preserving the statistics of AC power flow, which is critical for training GFMs. This preservation of the power flow statistics stands in contrast to approaches that perturb the admittance matrix directly or inject noise into the voltage outputs, both of which distort the underlying physics. Concretely, we derive $(\varepsilon,\delta)$-DP guarantees for the released voltage trajectories with respect to the admittance matrix, meaning privacy of the network parameters is obtained without any additional noise mechanism. Our bound depends on the adjacency assumption, the Jacobian of the AC power flow, and the covariance of the synthetic DP-loads. Finally, we present a synthetic voltage generation procedure and an empirical evaluation against Gaussian output-perturbation baselines, demonstrating that our approach provides a clear advantage for enabling GFM training.

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

The paper transfers (ε,δ)-DP from synthetic loads to voltage phasors via the true power flow map without extra noise, preserving physics better than output perturbation, but the bound's global validity is unclear due to local Jacobian linearization.

read the letter

The main point is that this work gets differential privacy on released voltage trajectories with respect to the admittance matrix by running DP synthetic loads through the real AC power flow equations. No separate noise is added to the voltages or the Ybus itself. That construction is new relative to the cited grid DP literature and lets the data keep the actual power flow statistics intact, which matters for training Grid Foundation Models on realistic behavior rather than distorted versions. The empirical comparison to Gaussian output-perturbation baselines shows a utility edge for their method, and the overall approach avoids the physics-breaking side effects of direct matrix or output perturbation. That part is done cleanly and is worth crediting. The soft spot is the privacy transfer itself. The bound is built from the load covariance, an adjacency assumption on neighboring Ybus datasets, and the power-flow Jacobian. Because AC power flow is nonlinear, the Jacobian only approximates sensitivity locally around one operating point. Without an explicit global Lipschitz bound or checks that the synthetic load distribution stays where the linearization error stays small, the actual privacy loss for some neighboring network pairs could exceed the claimed (ε,δ) under heavy loading or near voltage limits. The paper should clarify whether the derivation accounts for Jacobian variation or if the bound is only local. The adjacency definition also needs to be spelled out so it matches realistic network-parameter changes that the load covariance can actually mask. This paper is for power-systems researchers and ML groups working on privacy-preserving grid data. Readers focused on DP for physical simulations or synthetic data generation would get concrete value from the construction and the baseline comparison. It deserves peer review. The core idea is fresh enough and the experiments are there, so referees can verify the math and tighten the bound if needed.

Referee Report

3 major / 2 minor

Summary. The paper proposes releasing synthetic voltage phasor trajectories for distribution grids under differential privacy by first fitting a DP generative model to historical customer loads and then propagating the resulting synthetic load trajectories through the true AC power flow equations on the actual admittance matrix. The central claim is that the randomness already present in the DP synthetic loads suffices to obtain (ε,δ)-DP guarantees for the released voltages with respect to the admittance matrix itself, without any additional noise mechanism. The bound is stated to depend on an adjacency assumption for neighboring Ybus datasets, the Jacobian of the AC power flow, and the covariance of the synthetic DP-loads. This approach is argued to preserve the statistics of AC power flow (unlike direct perturbation of Ybus or voltage outputs) and is evaluated empirically against Gaussian output-perturbation baselines for enabling Grid Foundation Model training.

Significance. If the claimed DP transfer holds, the method would allow utilities to share realistic voltage trajectory corpora that respect the underlying nonlinear power-flow physics while simultaneously protecting both load profiles and network parameters. This is a potentially valuable contribution for privacy-preserving data release in power systems, as it avoids the distortion of physics-based statistics that occurs when noise is added directly to voltages or the admittance matrix. The approach could meaningfully support training of data-hungry grid ML models.

major comments (3)

[Abstract / DP guarantee derivation] Abstract (central claim): the (ε,δ)-DP bound for voltage trajectories w.r.t. the admittance matrix is derived by propagating DP synthetic loads through the AC power flow map and invoking the Jacobian. Because the AC power flow equations are nonlinear, the Jacobian supplies only a first-order local linearization around a specific operating point; without an explicit global Lipschitz constant or uniform bound on Jacobian variation over the support of the synthetic load distribution, the privacy loss for distant Ybus pairs could exceed the claimed (ε,δ) bound in regimes where linearization error is large (e.g., near voltage limits or heavy loading).
[Abstract / DP guarantee derivation] Abstract (adjacency assumption): the bound is stated to depend on 'the adjacency assumption' for neighboring Ybus datasets, yet the precise definition of neighboring admittance matrices (e.g., which realistic network-parameter changes are considered adjacent) is not visible. This definition is load-bearing because the load covariance must be shown to mask the output difference induced by any such neighboring pair; without it, the transfer of the DP guarantee cannot be verified.
[Abstract / Empirical evaluation] Abstract (empirical validation): the paper presents an empirical comparison against Gaussian output-perturbation baselines, but no direct empirical check or tightness assessment of the derived (ε,δ) bound itself (e.g., via privacy-loss histograms or worst-case neighboring Ybus pairs) is described. Post-hoc fitting of the generative-model covariance inside the load DP mechanism could affect the claimed tightness of the voltage bound.

minor comments (2)

The notation for the adjacency relation on Ybus and the precise dependence of the bound on the load covariance should be stated more explicitly to allow independent verification.
A short discussion of how the operating point for the Jacobian is chosen (or whether multiple linearizations are used) would improve clarity.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. These points help us strengthen the presentation of the DP transfer result. We respond to each major comment below and indicate the corresponding revisions.

read point-by-point responses

Referee: [Abstract / DP guarantee derivation] Abstract (central claim): the (ε,δ)-DP bound for voltage trajectories w.r.t. the admittance matrix is derived by propagating DP synthetic loads through the AC power flow map and invoking the Jacobian. Because the AC power flow equations are nonlinear, the Jacobian supplies only a first-order local linearization around a specific operating point; without an explicit global Lipschitz constant or uniform bound on Jacobian variation over the support of the synthetic load distribution, the privacy loss for distant Ybus pairs could exceed the claimed (ε,δ) bound in regimes where linearization error is large (e.g., near voltage limits or heavy loading).

Authors: We agree that the AC power flow map is nonlinear and that the Jacobian provides only a local linearization. In the revision we will add an explicit statement that the derived bound is valid under the assumption of load variations remaining within a neighborhood of the linearization point (quantified via the synthetic load covariance). We will also include a short analysis of the linearization error using the second-order remainder term of the power-flow map and numerical checks over the support of the DP load distribution to confirm that the error remains negligible for the operating regimes considered in the paper. A global Lipschitz constant is not required under these conditions and would be overly conservative; we will note this limitation clearly. revision: partial
Referee: [Abstract / DP guarantee derivation] Abstract (adjacency assumption): the bound is stated to depend on 'the adjacency assumption' for neighboring Ybus datasets, yet the precise definition of neighboring admittance matrices (e.g., which realistic network-parameter changes are considered adjacent) is not visible. This definition is load-bearing because the load covariance must be shown to mask the output difference induced by any such neighboring pair; without it, the transfer of the DP guarantee cannot be verified.

Authors: The adjacency relation is defined as two Ybus matrices whose difference is bounded in a chosen matrix norm (e.g., maximum absolute entry-wise difference) and corresponds to realistic parameter perturbations such as line-impedance variations of at most 10 %. We will insert a formal definition of this adjacency relation together with the precise norm and the resulting bound on the voltage difference induced by the power-flow map. This will make explicit how the covariance of the DP synthetic loads is sufficient to mask the output difference and thereby transfer the (ε,δ) guarantee. revision: yes
Referee: [Abstract / Empirical evaluation] Abstract (empirical validation): the paper presents an empirical comparison against Gaussian output-perturbation baselines, but no direct empirical check or tightness assessment of the derived (ε,δ) bound itself (e.g., via privacy-loss histograms or worst-case neighboring Ybus pairs) is described. Post-hoc fitting of the generative-model covariance inside the load DP mechanism could affect the claimed tightness of the voltage bound.

Authors: We accept that a direct empirical assessment of the privacy bound would improve the paper. In the revised version we will add a new subsection that (i) constructs worst-case neighboring Ybus pairs consistent with the adjacency definition, (ii) evaluates the empirical privacy loss for the released voltages, and (iii) presents privacy-loss histograms. We will also clarify that the covariance used in the bound is obtained from the DP generative model itself and that the bound is stated conservatively with respect to any post-hoc fitting. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the DP bound derivation

full rationale

The paper derives (ε,δ)-DP for voltage trajectories w.r.t. the admittance matrix by propagating DP synthetic loads through the AC power flow map. The bound is explicitly stated to depend on an adjacency assumption for neighboring Ybus, the Jacobian, and the covariance of the DP loads. No equations or steps are shown that reduce the claimed guarantee to a tautology, a fitted parameter renamed as prediction, or a self-citation chain. The use of upstream DP load noise to mask Ybus changes follows standard post-processing arguments under the given linearization; the covariance enters as an external property of the load mechanism rather than being redefined by the voltage result. No self-definitional, ansatz-smuggling, or renaming patterns appear in the provided text.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard differential privacy definitions, the AC power flow equations, and the assumption that the Jacobian and load covariance allow privacy transfer. No new physical entities are postulated. The generative model for loads introduces fitted parameters whose covariance enters the bound.

free parameters (2)

DP privacy budget (ε, δ)
Chosen for the load generative model; directly affects the covariance term in the voltage bound.
Covariance of synthetic DP-loads
Obtained from the fitted DP generative model on historical loads; appears in the derived privacy bound for voltages.

axioms (2)

domain assumption AC power flow equations map loads and admittance matrix to voltage phasors.
Used when propagating synthetic loads to voltages; standard in power systems but treated as exact here.
standard math Differential privacy definition with respect to an adjacency relation on the admittance matrix.
The bound is derived under a specific adjacency assumption for neighboring network parameters.

pith-pipeline@v0.9.0 · 5588 in / 1417 out tokens · 63153 ms · 2026-05-08T19:00:59.672872+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

39 extracted references · 6 canonical work pages · 1 internal anchor

[1]

A survey on the use of synthetic data for enhancing key aspects of trustworthy ai in the energy domain: Challenges and opportunities,

M. Meiser and I. Zinnikus, “A survey on the use of synthetic data for enhancing key aspects of trustworthy ai in the energy domain: Challenges and opportunities,”Energies, vol. 17, no. 9, p. 1992, 2024

1992
[2]

Ai-based synthetic data generation tech- niques for improved fault classification in power systems,

P. K. Shukla and K. Deepa, “Ai-based synthetic data generation tech- niques for improved fault classification in power systems,”Ain Shams Engineering Journal, vol. 16, no. 8, p. 103485, 2025

2025
[3]

Foundation models for the electric power grid,

H. F. Hamann, T. Brunschwiler, B. Gjorgiev, L. S. A. Martins, A. Puech, A. Varbella, J. Weiss, J. Bernabe-Moreno, A. Blondin Mass ´e, S. Choi, I. Foster, B.-M. Hodge, R. Jain, K. Kim, V . Mai, F. Mirall `es, M. De Montigny, O. Ramos-Lea ˜nos, H. Supr ˆeme, L. Xie, E.-N. S. Youssef, A. Zinflou, A. J. Belyi, R. J. Bessa, B. P. Bhattarai, J. Schmude, and S. ...

2024
[4]

Topology estimation using graphical models in multi-phase power distribution grids,

D. Deka, M. Chertkov, and S. Backhaus, “Topology estimation using graphical models in multi-phase power distribution grids,”IEEE Trans- actions on Power Systems, vol. 35, no. 3, pp. 1663–1673, 2020

2020
[5]

Unbalanced multi-phase distribution grid topology estimation and bus phase identification,

Y . Liao, Y . Weng, G. Liu, Z. Zhao, C. W. Tan, and R. Rajagopal, “Unbalanced multi-phase distribution grid topology estimation and bus phase identification,”IET Smart Grid, vol. 2, no. 4, pp. 557–570, 2019. 11

2019
[6]

Learning distribution grid topologies: A tutorial,

D. Deka, V . Kekatos, and G. Cavraro, “Learning distribution grid topologies: A tutorial,”IEEE Transactions on Smart Grid, vol. 15, no. 1, pp. 999–1013, 2024

2024
[7]

Inverse power flow problem,

Y . Yuan, S. H. Low, O. Ardakanian, and C. J. Tomlin, “Inverse power flow problem,”IEEE Transactions on Control of Network Systems, vol. 10, no. 1, pp. 261–273, 2023

2023
[8]

Privacy-preserving obfuscation for distributed power systems,

T. W. Mak, F. Fioretto, and P. Van Hentenryck, “Privacy-preserving obfuscation for distributed power systems,”Electric Power Systems Research, vol. 189, p. 106718, 2020

2020
[9]

Differentially private state estimation in distribution networks with smart meters,

H. Sandberg, G. D ´an, and R. Thobaben, “Differentially private state estimation in distribution networks with smart meters,” in2015 54th IEEE conference on decision and control (CDC), pp. 4492–4498, IEEE, 2015

2015
[10]

A comprehensive survey on privacy-preserving technologies for smart grids,

H. Bibi, M. Abolhasan, J. Lipman, M. Abdollahi, and W. Ni, “A comprehensive survey on privacy-preserving technologies for smart grids,”Computers and Electrical Engineering, vol. 124, p. 110371, 2025

2025
[11]

A large synthetic dataset for machine learning applications in power transmission grids,

M. Gillioz, G. Dubuis, and P. Jacquod, “A large synthetic dataset for machine learning applications in power transmission grids,”Scientific Data, vol. 12, no. 1, p. 168, 2025

2025
[12]

Generating synthetic energy time series: A review,

M. Turowski, B. Heidrich, L. Weing ¨artner, L. Springer, K. Phipps, B. Sch¨afer, R. Mikut, and V . Hagenmeyer, “Generating synthetic energy time series: A review,”Renewable and Sustainable Energy Reviews, vol. 206, p. 114842, 2024

2024
[13]

Gen- erative adversarial network for synthetic time series data generation in smart grids,

C. Zhang, S. R. Kuppannagari, R. Kannan, and V . K. Prasanna, “Gen- erative adversarial network for synthetic time series data generation in smart grids,” in2018 IEEE international conference on communications, control, and computing technologies for smart grids (SmartGridComm), pp. 1–6, IEEE, 2018

2018
[14]

The algorithmic foundations of differential privacy,

C. Dwork, A. Roth,et al., “The algorithmic foundations of differential privacy,”Foundations and Trends® in Theoretical Computer Science, vol. 9, no. 3–4, pp. 211–407, 2014

2014
[15]

Generating synthetic data with formal privacy guarantees: State of the art and the road ahead,

V . Schlegel, A. A. Bharath, Z. Zhao, and K. Yee, “Generating synthetic data with formal privacy guarantees: State of the art and the road ahead,” arXiv preprint arXiv:2503.20846, 2025

work page arXiv 2025
[16]

I have a dream! (differentially private smart metering),

G. ´Acs and C. Castelluccia, “I have a dream! (differentially private smart metering),” inInformation Hiding(T. Filler, T. Pevn ´y, S. Craver, and A. Ker, eds.), (Berlin, Heidelberg), pp. 118–132, Springer Berlin Heidelberg, 2011

2011
[17]

Differential privacy for real smart metering data,

G. Eibl and D. Engel, “Differential privacy for real smart metering data,” Computer Science-Research and Development, vol. 32, no. 1, pp. 173– 182, 2017

2017
[18]

Lightweight privacy for smart metering data by adding noise,

P. Barbosa, A. Brito, H. Almeida, and S. Claus, “Lightweight privacy for smart metering data by adding noise,” inProceedings of the 29th Annual ACM Symposium on Applied Computing, pp. 531–538, 2014

2014
[19]

Deep learning with differential privacy,

M. Abadi, A. Chu, I. Goodfellow, H. B. McMahan, I. Mironov, K. Talwar, and L. Zhang, “Deep learning with differential privacy,” in Proceedings of the 2016 ACM SIGSAC conference on computer and communications security, pp. 308–318, 2016

2016
[20]

Dpwgan: High-quality load profiles synthesis with differential privacy guarantees,

J. Huang, Q. Huang, G. Mou, and C. Wu, “Dpwgan: High-quality load profiles synthesis with differential privacy guarantees,”IEEE Transac- tions on Smart Grid, vol. 14, no. 4, pp. 3283–3295, 2023

2023
[21]

Constrained

M. Hoseinpour and V . Dvorkin, “Constrained diffusion models for synthesizing representative power flow datasets,”arXiv preprint arXiv:2506.11281, 2025

work page arXiv 2025
[22]

Differential privacy for power grid obfuscation,

F. Fioretto, T. W. K. Mak, and P. Van Hentenryck, “Differential privacy for power grid obfuscation,”IEEE Transactions on Smart Grid, vol. 11, no. 2, pp. 1356–1366, 2020

2020
[23]

Differentially private algorithms for synthetic power system datasets,

V . Dvorkin and A. Botterud, “Differentially private algorithms for synthetic power system datasets,”IEEE Control Systems Letters, vol. 7, pp. 2053–2058, 2023

2053
[24]

Differentially private k-means cluster- ing applied to meter data analysis and synthesis,

N. Ravi, A. Scaglione, S. Kadam, R. Gentz, S. Peisert, B. Lunghino, E. Levijarvi, and A. Shumavon, “Differentially private k-means cluster- ing applied to meter data analysis and synthesis,”IEEE transactions on smart grid, vol. 13, no. 6, pp. 4801–4814, 2022

2022
[25]

Differentially private distribution release of gaussian mixture models via kl-divergence minimization,

H. Liu, A. Scaglione, and S. Peisert, “Differentially private distribution release of gaussian mixture models via kl-divergence minimization,” arXiv preprint arXiv:2506.03467, 2025

work page arXiv 2025
[26]

Privacy-preserving federated learning for residential short-term load forecasting,

J. D. Fern ´andez, S. P. Menci, C. M. Lee, A. Rieger, and G. Fridgen, “Privacy-preserving federated learning for residential short-term load forecasting,”Applied energy, vol. 326, p. 119915, 2022

2022
[27]

Realistic differentially-private transmission power flow data release,

D. Smith, F. Geth, E. Vercoe, A. Feutrill, M. Ding, J. Chan, J. Foster, and T. Rakotoarivelo, “Realistic differentially-private transmission power flow data release,”arXiv preprint arXiv:2103.14036, 2021

work page arXiv 2021
[28]

Differentially private optimal power flow for distribution grids,

V . Dvorkin, F. Fioretto, P. Van Hentenryck, P. Pinson, and J. Kazempour, “Differentially private optimal power flow for distribution grids,”IEEE Transactions on Power Systems, vol. 36, no. 3, pp. 2186–2196, 2020

2020
[29]

Privacy- preserving power system obfuscation: A bilevel optimization approach,

T. W. K. Mak, F. Fioretto, L. Shi, and P. Van Hentenryck, “Privacy- preserving power system obfuscation: A bilevel optimization approach,” IEEE Transactions on Power Systems, vol. 35, no. 2, pp. 1627–1637, 2020

2020
[30]

The influence of differential privacy on short term electric load forecasting,

G. Eibl, K. Bao, P.-W. Grassal, D. Bernau, and H. Schmeck, “The influence of differential privacy on short term electric load forecasting,” Energy Informatics, vol. 1, no. Suppl 1, p. 48, 2018

2018
[31]

Characterizing probability density distributions for household electricity load profiles from high- resolution electricity use data,

J. Munkhammar, J. Ryd ´en, and J. Wid ´en, “Characterizing probability density distributions for household electricity load profiles from high- resolution electricity use data,”Applied Energy, vol. 135, pp. 382–390, 2014

2014
[32]

Prediction of energy consumption for variable customer portfolios including aleatoric uncertainty estimation,

O. Mey, A. Schneider, O. Enge-Rosenblatt, Y . Bravo, and P. Stenzel, “Prediction of energy consumption for variable customer portfolios including aleatoric uncertainty estimation,” in2021 10th International Conference on Power Science and Engineering (ICPSE), pp. 61–71, IEEE, 2021

2021
[33]

Practical lognormal framework for household energy consumption modeling,

P. Kuusela, I. Norros, R. Weiss, and T. Sorasalmi, “Practical lognormal framework for household energy consumption modeling,”Energy and Buildings, vol. 108, pp. 223–235, 2015

2015
[34]

SoK: Practical Aspects of Releasing Differentially Private Graphs

N. D’Silva, S. Nepal, and S. S. Kanhere, “Sok: Practical aspects of re- leasing differentially private graphs,”arXiv preprint arXiv:2603.18779, 2026

work page internal anchor Pith review Pith/arXiv arXiv 2026
[35]

Privacy: Theory meets practice on the map,

A. Machanavajjhala, D. Kifer, J. Abowd, J. Gehrke, and L. Vilhuber, “Privacy: Theory meets practice on the map,” in2008 IEEE 24th international conference on data engineering, pp. 277–286, IEEE, 2008

2008
[36]

Generalization error bounds using wasserstein distances,

A. T. Lopez and V . Jog, “Generalization error bounds using wasserstein distances,” in2018 ieee information theory workshop (ITW), pp. 1–5, IEEE, 2018

2018
[37]

Sample IEEE123 bus system for OEDI SI

T. Elgindy and K. Balasubramaniam, “Sample IEEE123 bus system for OEDI SI.” Open Energy Data Initiative (OEDI), National Renewable Energy Laboratory, Dataset, 2022. doi: 10.25984/2228282, accessed: 2026-04-16

work page doi:10.25984/2228282 2022
[38]

Concentrated differential privacy: Simpli- fications, extensions, and lower bounds,

M. Bun and T. Steinke, “Concentrated differential privacy: Simpli- fications, extensions, and lower bounds,” inTheory of cryptography conference, pp. 635–658, Springer, 2016

2016
[39]

Adaptive estimation of a quadratic functional by model selection,

B. Laurent and P. Massart, “Adaptive estimation of a quadratic functional by model selection,”Annals of statistics, pp. 1302–1338, 2000. APPENDIX A. Supporting Lemma Lemma 1(Log is Lipschitz on a positive interval).Ifx, x ′ ∈ [pmin, pmax]withp min >0, then|logx−logx ′| ≤ |x− x′|/pmin.Proof.By the mean value theorem,logx−logx ′ = (x−x ′)/ξfor someξ∈[p min,...

2000