Pith. sign in

REVIEW 3 major objections 5 minor 35 references

Statistical Disaggregation -- a Monte Carlo Approach for Imputation under Constraints

T0 review · 3 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read This paper introduces the Constrained Fusion sampler, which returns exact i.i.d.

desk verdict A genuinely new and mostly sound exact sampler for linear equality constraints; the nonlinear extension is advertised too strongly and, as printed, one acceptance line in Algorithm 2 appears to have a missing minus sign. read the letter →

arxiv 2504.18377 v1 pith:BJMMWQ42 submitted 2025-04-25 stat.CO

Shenggang Hu , Hongsheng Dai , Fanlin Meng , Louis Aslett , Murray Pollock , Gareth O. Roberts This is my paper · ORCID
classification stat.CO
keywords statisticaldisaggregationequalityconstraintsexactsamplingLangevindiffusionrejectionBrownianbridgetimeseriesimputationconstrainedMonteCarlo
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

This paper tackles a common but intractable statistical task: given several independent component densities $f_1,\dots,f_m$ and an equality constraint that ties their values together, draw samples from the joint density restricted to the constraint. Standard rejection sampling fails because the constraint has zero Lebesgue measure, and Markov chain methods on the constraint can miss disconnected modes. The authors' Constrained Fusion sampler instead couples the components through Brownian bridges whose endpoints are forced to satisfy the constraint, then applies an exact path-space rejection step. If the constraint is linear, the method returns exact independent samples; on nonlinear constraints it reduces the problem to drawing uniform samples from the constraint manifold, and the simulations show it can discover all modes of a multimodal target where a constrained Hamiltonian sampler gets stuck. A sympathetic reader would care because this makes disaggregation and imputation problems, such as recovering three daily electricity readings from a single daily meter total, statistically principled and gives calibrated uncertainty intervals.

What carries the argument

The engine is the augmented target–proposal pair in Proposition 1: target $g_H(x^{(1:m)}, y^{(1:m)})\propto \prod_i f_i^2(x^{(i)}) p_i(y^{(i)}|x^{(i)})/f_i(y^{(i)})$ on $\mathbb{R}^{md}\times H$ and proposal $h_H\propto \prod_i f_i(x^{(i)}) \cdot N(y^{(i)};x^{(i)},T I_d)$ restricted to $H$. Their Radon–Nikodym derivative, Corollary 3, is the expectation over Brownian bridges of $\exp(-\sum_i \int_0^T \phi_i(\omega_s^{(i)})ds)$ with $\phi_i(u)=\frac12(\|\nabla\log f_i(u)\|^2+\nabla\cdot\nabla\log f_i(u))-l_i\ge 0$, so rejection sampling on this derivative corrects any constrained-Gaussian proposal into the target. The second key piece is that the constrained Gaussian endpoint distribution is tractable: under a linear constraint $Ay=c$ the endpoints are Gaussian with an explicit normalising constant $Z_H(x^{(1:m)})\propto \exp(-(c-Ax)^\top(AA^\top)^{-1}(c-Ax)/(2T))$, which gives the first-stage acceptance probability. For nonlinear constraints the same rejection identity holds if endpoints are drawn uniformly from $H$, decoupling the distribution from the geometry of the constraint.

What would settle it

Run Algorithm 2 on a nonlinear constraint where the prescribed uniform sampler provably fails to mix within a fixed budget—say a thin curved constraint connecting two distant lobes—and compare the output with ground truth from independent long-run MCMC or numerical integration. If the Constrained Fusion output fails to place mass on both lobes, the claim that it is exact on arbitrary constraints is falsified in that regime; if it still succeeds, the inner sampler's non-convergence turns out to be harmless.

Watch

Extended reading notes

Core claim

The central claim is that sampling from $f_H(y^{(1:m)})\propto \prod_i f_i(y^{(i)}) \mathbb{I}_{y^{(1:m)}\in H}$ can be done exactly by simulating $m$ independent Langevin diffusions whose invariant laws are $f_i^2$, drawing Brownian-bridge proposals from initial points $x^{(i)}\sim f_i$ to endpoints $y^{(i)}$ on $H$ distributed as a constrained Gaussian (or uniform, in the general case), and accepting with probability given by the expectation of $\exp(-\sum_i \int_0^T \phi_i(\omega_s^{(i)})ds)$ over Brownian bridges. Proposition 1 identifies this proposal as a valid augmented distribution whose $y$-marginal is exactly $f_H$, and the rejection step is implemented exactly through Poisson-thinning when the $\phi_i$ are bounded, with a layered Brownian-bridge construction for unbounded $\phi_i$. For linear constraints Algorithm 1 is exact; for arbitrary smooth manifolds Algorithm 2 is exact provided one can sample uniformly from $H$.

Load-bearing premise

The load-bearing premise is that, for nonlinear constraints, the step that draws uniform points from the constraint surface can be done exactly or at least mixed well enough; the paper does not prove this for its inner sampler, so if that sampler misses parts of the surface the whole exact-sampling claim collapses.

Editorial extensions

If this is right

  • For any linear constraint, Algorithm 1 produces exact i.i.d. samples from the constrained product density, so the returned samples need no burn-in or convergence diagnostics.
  • Adding a sum constraint always reduces the total variance of a Gaussian model, and when model uncertainty exceeds its bias the constrained imputation also has lower mean squared error; simulations indicate the same holds for heavy-tailed non-Gaussian models.
  • The method turns time-series disaggregation into sequential constrained sampling: imputing high-resolution readings from low-frequency totals gives tighter, non-inflating uncertainty bands than unconstrained forecasting.
  • On nonlinear constraints such as mean-plus-variance constraints, the algorithm explores multiple modes of the constrained distribution with far fewer samples than a constrained Hamiltonian sampler, because the distribution and the constraint are handled separately.
  • Only one tuning parameter $T$ affects efficiency, not accuracy; the user does not have to pick step sizes, mass matrices or proposal scales for the target itself.

Reading between the lines

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

  • The advertised exactness on arbitrary nonlinear constraints inherits fully from the assumption that the inner uniform sampler on $H$ mixes; because the paper gives no convergence guarantee for that step, in practice the nonlinear version is exact only up to the quality of the CHMC uniform samples.
  • An immediate testable extension is to swap the CHMC uniform sampler for a different uniform-on-manifold sampler, or an exact sampler when one exists, and check whether the fused output remains correct; this would isolate where the nonlinear method's error enters.
  • The MSE analysis suggests a practical decision rule for imputation: impose the constraint when residual variance dominates bias, since otherwise per-component errors can grow even when total variance shrinks — a caveat for naive users of constrained imputation.
  • Because the acceptance probability decays when the constraint is far from the typical set of the product density, a sequential Monte Carlo version of the fusion step — analogous to the paper's own proposed future work — is the natural route to scaling the method to high-dimensional or far-away constraints.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The manuscript introduces the Constrained Fusion sampler, a rejection sampling method for equality-constrained product densities of the form f_H(y) ∝ ∏_{i=1}^m f_i(y_i) 1_{h(y)=0}. The method represents each f_i as the invariant distribution of a Langevin diffusion, proposes Brownian-bridge endpoints that satisfy the constraint (for linear constraints via a constrained Gaussian, and for general constraints via uniform samples from the manifold H), and applies a path-space rejection step based on the Girsanov density. The paper proves the acceptance ratio in Lemma 2 and Corollary 3, gives two algorithms, compares against importance sampling, random-walk Metropolis-Hastings, and CHMC in simulation studies, and applies the method to disaggregation of electricity consumption and medicine price data.

Significance. The core linear-constraint exactness claim is well supported: the derivation of the acceptance ratio is coherent, the algorithm produces i.i.d. samples when its ingredients are exact, and the simulation studies demonstrate practical gains in mean and variance estimation and in exploring the four modes of a circle-constrained target. The decoupling of the target distribution from the constraint is an elegant idea with potential value in disaggregation and imputation. The paper is considerably weaker for the advertised 'arbitrary constraints' claim because Algorithm 2 inherits correctness from an unverified uniform sampler on H; this part needs substantial additional support or a careful re-scoping of the claims.

major comments (3)
  1. [Section 3.2 (Case 2), Algorithm 2, Appendix B.5] The exactness of the arbitrary-constraint sampler requires y to be drawn exactly from U(H), the uniform distribution on the constraint manifold. The implementation uses CHMC as the inner sampler, but no convergence guarantee, burn-in, thinning, or stationarity diagnostics are reported for any of the nonlinear experiments (Fig. 3, Fig. 10, and Study 3 in Appendix E.3). If the inner chain has not reached stationarity, the proposal is some Q ≠ U(H), and the rejection ratio in Steps 4–8 of Algorithm 2 is no longer the Radon–Nikodym ratio of Corollary 3; accepted draws are then not i.i.d. from f_H, and the claimed natural handling of multimodal targets on arbitrary constraints is not established. This is load-bearing for the second half of the paper's central claim. Please provide verifiable diagnostics or mixing guarantees for the specific manifolds used, or explicitly state that the nonlinear algorithm is approximate.
  2. [Section 4.2, Study 1, display of H_t] The nonlinear constraint is written as ∑_i Y_i^{(t)} = S_t and ∑_i (Y_i^{(t)} − S_t)^2 = Σ_t. If S_t is the daily total, the variance of the three within-day readings is proportional to ∑_i (Y_i^{(t)} − S_t/3)^2, so the displayed constraint is not the 'variance constraint' described in the text. Please correct the formula and confirm which constraint was used for Figure 6b; the interpretation of the uncertainty comparisons in Study 1 depends on this.
  3. [Section 3.1, Lemma 2 and Condition 1] The exactness of the rejection step is conditional on Condition 1 holding for every f_i, but the paper does not verify the condition for the Student-t and Generalized Logistic densities used in the experiments, and it does not state explicitly whether the exact layered Brownian bridge construction of Appendix B.2 was used when φ_i is unbounded. Please add these verifications or state the additional assumptions; otherwise the 'exact' label on the simulation results is not fully supported.
minor comments (5)
  1. [Algorithm 2, Step 4] The printed acceptance condition `log U1 ≤ ∥y−x∥^2/(2T)` is missing the minus sign; it should be `log U1 ≤ −∥y−x∥^2/(2T)` or equivalently `U1 ≤ exp(−∥y−x∥^2/(2T))`.
  2. [Equations (6) and (7)] The constants l_i disappear silently between equation (6) and equation (7); please state explicitly that they are absorbed into the proportionality constant of the Radon–Nikodym ratio.
  3. [Appendix F, Proposition 5] The displayed identity should read tr(Σ*) − ∑ σ_i^2 = −(1/w^2) ∑ σ_i^4; as written the sign is wrong, although the stated reduction is positive.
  4. [Appendix E.2, Definition 6] The dimension of the FGM copula parameter θ is stated as R^{2m−m−1}; the standard dimension for an m-variate FGM copula is 2^m − m − 1.
  5. [Appendix E.2, Study 2] Study 2 is not a product-density example; the text should explain how the single-target-density formulation maps onto Proposition 1 and the algorithm for product densities.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the Constrained Fusion sampler is derived constructively from a diffusion-bridge augmentation identity and Girsanov-based rejection, and the application predictions are evaluated on held-out data.

full rationale

The paper derives its sampler from first principles rather than from fitted inputs or self-citation loops. The target f_H is not assumed; Proposition 1 defines the augmented density g_H and verifies by integration that its marginal over endpoints is the product density restricted to H. Lemma 2 and Corollary 3 obtain the rejection weight from the Girsanov formula for Langevin bridges, using published results (Hansen et al. 2003; Beskos et al. 2006, 2008) whose assumptions do not include the present target or algorithm output. Algorithms 1 and 2 are constructive rejection samplers whose inputs are the component densities f_i, the constraint H, and the tunable bridge length T; T is stated to affect efficiency rather than accuracy, and Algorithm 2 explicitly conditions on the ability to sample uniformly from H. The real-data studies fit AR/Generalized-Logistic parameters on a training set and impute left-out households/days, so the constrained estimates are genuine predictions rather than re-statements of fitted values. Self-citations to earlier fusion work are background and extension, not load-bearing support for exactness. The CHMC convergence caveat in Algorithm 2 and Appendix B.5 is a correctness/ergodicity risk for the nonlinear-constraint case, not a circularity of the derivation.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central sampling algorithm introduces no new physical or mathematical entities. Its free parameters are limited to tuning choices (T, K, regularization weights) and the application-specific fitted GenLog parameters. The load-bearing assumptions are standard diffusion theory plus the practical requirement that the rejection event can be simulated exactly and that nonlinear constraints are sampled uniformly.

free parameters (3)
  • Diffusion time T = chosen by hand per experiment
    T controls the Brownian bridge length and the acceptance probability; the paper states the authors used 'an appropriate value of T' for each simulation (Section 3.3). It affects efficiency, not the target distribution.
  • AR order K = 7
    Set to 7 because weekly cycles are assumed for household electricity use (Section 4.2). This is a modeling choice, not an algorithm parameter.
  • GenLog fitting regularizers lambda1, lambda2 = lambda1 = 1e-3, lambda2 = 1e-6
    Chosen by hand in Appendix D, Remark 11, to stabilize moment matching of Generalized Logistic parameters. These affect the empirical model, not the sampler derivation.
assumptions (5)
  • standard math Regular Value Theorem for the constraint set H
    Appendix A, Lemma 4, requires h smooth with surjective derivative so that H is a manifold with a canonical volume measure. This justifies the dominating measure in (1).
  • standard math Girsanov theorem and diffusion bridge Radon-Nikodym derivative
    Lemma 2 relies on the Beskos et al. 2006 factorisation of the diffusion measure and the martingale condition in Condition 1 to express the transition density pi(y|x) in closed form.
  • domain assumption Boundedness or exact layered simulation of phi_i
    The rejection event in (7) can be simulated exactly only if each phi_i is bounded above or the layered Brownian bridge construction of Beskos et al. 2008 is implemented. The paper assumes this for the distributions used, but does not verify it for the t-distribution examples.
  • domain assumption Independent components conditioned on history
    The time series model in Section 4.1 assumes Y_t^(i) are independent across i given the past, matching the product structure of (1). The authors note a copula extension in Remark 5 and Appendix E.2.
  • ad hoc to paper Well-mixed uniform sampler on nonlinear constraint manifolds
    Algorithm 2 for nonlinear constraints requires uniform samples from H, generated by CHMC. The exactness of the whole algorithm is conditional on this inner sampler having converged, which is not proven.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Statistical Disaggregation -- a Monte Carlo Approach for Imputation under Constraints." pith.science (2026). https://pith.science/paper/BJMMWQ42

@misc{pith2026250418377,
  author       = {Pith},
  title        = {Pith review of: Statistical Disaggregation -- a Monte Carlo Approach for Imputation under Constraints},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BJMMWQ42}},
  note         = {Machine review of arXiv:2504.18377}
}
read the original abstract

Equality-constrained models naturally arise in problems in which measurements are taken at different levels of resolution. The challenge in this setting is that the models usually induce a joint distribution which is intractable. Resorting to instead sampling from the joint distribution by means of a Monte Carlo approach is also challenging. For example, a naive rejection sampling does not work when the probability mass of the constraint is zero. A typical example of such constrained problems is to learn energy consumption for a higher resolution level based on data at a lower resolution, e.g., to decompose a daily reading into readings at a finer level. We introduce a novel Monte Carlo sampling algorithm based on Langevin diffusions and rejection sampling to solve the problem of sampling from equality-constrained models. Our method has the advantage of being exact for linear constraints and naturally deals with multimodal distributions on arbitrary constraints. We test our method on statistical disaggregation problems for electricity consumption datasets, and our approach provides better uncertainty estimation and accuracy in data imputation compared with other naive/unconstrained methods.

Figures

Figures reproduced from arXiv: 2504.18377 by the authors.

Figure 1
Figure 1. Residual distribution plot based on AR model (without constraints) for the Irish [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The percentage error curve varying with sample size [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Drawn samples using Constrained Fusion (left) and CHMC (right) plotted, the colour [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figures from the paper (7 more)
Figure 5
Figure 5. Figure 5: Constrained Imputation for Time series. By the construction of the AR model, every time point Yt depends only on its previous times Yt−1,Yt−2, . . . , and the constraint Yt ∈ Ht . To illustrate the workflow, we will assume a sum constraint, Pm i=1 Y (i) t = St , for no…
Figure 6
Figure 6. Figure 6: Energy consumption imputation and error with and without constraints [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: An example of the Poisson process rejection. The blue curve is the trajectory of [PITH_FULL_IMAGE:figures/full_fig_p024_7.png]
Figure 8
Figure 8. Figure 8: CDF and PDF of simulated data against ground truth [PITH_FULL_IMAGE:figures/full_fig_p030_8.png]
Figure 9
Figure 9. Figure 9: Percentage difference prediction of peak and trough values, comparing the constrained [PITH_FULL_IMAGE:figures/full_fig_p034_9.png]
Figure 10
Figure 10. Figure 10: Medicine price disaggregated given sample mean and covariance. [PITH_FULL_IMAGE:figures/full_fig_p035_10.png]
Figure 11
Figure 11. Figure 11: Improvements in accuracy when adding sum constraint for different cases. Improve [PITH_FULL_IMAGE:figures/full_fig_p039_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 34 canonical work pages

  1. [1]

    and Agrawal, R

    Adhikari, R. and Agrawal, R. K. (2013) An Introductory Study on Time Series Modeling and Forecasting . arXiv preprint arXiv:1302.6613

    arXiv 2013

  2. [2]

    and Bourotte, M

    Allard, D. and Bourotte, M. (2015) Disaggregating Daily Precipitations into Hourly Values with a Transformed Censored Latent Gaussian Process . Stochastic Environmental Research and Risk Assessment, 29, 453--462

    work page 2015

  3. [3]

    and Escher, J

    Amann, H. and Escher, J. (2009) Analysis III . Springer

    work page 2009

  4. [4]

    and Roberts, G

    Beskos, A., Papaspiliopoulos, O. and Roberts, G. O. (2008) A Factorisation of Diffusion Measure and Finite Sample Path Constructions . Methodology and Computing in Applied Probability, 10, 85--104

    work page 2008

  5. [5]

    Beskos, A., Papaspiliopoulos, O., Roberts, G. O. and Fearnhead, P. (2006) Exact and Computationally Efficient Likelihood-based Estimation for Discretely Observed Diffusion Processes (with discussion) . Journal of the Royal Statistical Society: Series B (Statistical Methodology), 68, 333--382

    work page 2006

  6. [6]

    and Urtasun, R

    Brubaker, M., Salzmann, M. and Urtasun, R. (2012) A Family of MCMC Methods on Implicitly Defined Manifolds . In Artificial Intelligence and Statistics, 161--172

    work page 2012

  7. [7]

    L., Moser, C., Ramseier, C., Samuel, R., Sander, D., Schmidt, S., Sohre, A

    Burger, P., Bezençon, V., Bornemann, B., Brosch, T., Carabias-Hütter, V., Farsi, M., Hille, S. L., Moser, C., Ramseier, C., Samuel, R., Sander, D., Schmidt, S., Sohre, A. and Volland, B. (2015) Advances in Understanding Energy Consumption Behavior and the Governance of Its Change – Outline of an Integrated Framework . Frontiers in Energy Research, https:/...

    work page arXiv 2015

  8. [8]

    and Girolami, M

    Byrne, S. and Girolami, M. (2013) Geodesic Monte Carlo on Embedded Manifolds . Scandinavian Journal of Statistics, 40, 825--845

    work page 2013

Show all 35 references
  1. [9]

    Chua, A. J. (2020) Sampling from Manifold-restricted Distributions Using Tangent Bundle Projections . Statistics and Computing, 30, 587--602

    2020

  2. [10]

    https://www.ucd.ie/issda/data/commissionforenergyregulationcer/

    Commission for Energy Regulation (CER) (2009-2010 a ) CER Smart Metering Project - Electricity Customer Behaviour Trial [dataset] . https://www.ucd.ie/issda/data/commissionforenergyregulationcer/. Irish Social Science Data Archive. 1st Edition, SN: 0012-00

    2009

  3. [11]

    https://www.ucd.ie/issda/data/commissionforenergyregulationcer/

    --- (2009-2010 b ) CER Smart Metering Project - Electricity Customer Behaviour Trial [dataset] . https://www.ucd.ie/issda/data/commissionforenergyregulationcer/. Irish Social Science Data Archive. 1st Edition, SN: 0013-00

    2009

  4. [12]

    Cong, Y., Chen, B., Zhou, M. et al. (2017) Fast Simulation of Hyperplane-truncated Multivariate Normal Distributions . Bayesian Analysis, 12, 1017--1037

    2017

  5. [13]

    (2017) A New Rejection Sampling Method Without Using Hat Function

    Dai, H. (2017) A New Rejection Sampling Method Without Using Hat Function . Bernoulli, 23, 2434--2465

    2017

  6. [14]

    and Roberts, G

    Dai, H., Pollock, M. and Roberts, G. (2019) Monte Carlo Fusion . Journal of Applied Probability, 56, 174--191

    2019

  7. [15]

    and Roberts, G

    Dai, H., Pollock, M. and Roberts, G. O. (2023) Bayesian fusion: Scalable unification of distributed statistical analyses. Journal of the Royal Statistical Society Series B: Statistical Methodology, 85, 84--107

    2023

  8. [16]

    and S \"a rndal, C.-E

    Deville, J.-C. and S \"a rndal, C.-E. (1992) Calibration estimators in survey sampling. Journal of the American statistical Association, 87, 376--382

    1992

  9. [17]

    M., Hughes, J., Vats, D., Dai, N., Gupta, K

    Flegal, J. M., Hughes, J., Vats, D., Dai, N., Gupta, K. and Maji, U. (2021) mcmcse: Monte Carlo Standard Errors for MCMC. Riverside, CA, and Kanpur, India. R package version 1.5-0

    2021

  10. [18]

    Halliwell, L. J. (2018) The log-gamma distribution and non-normal error. Variance (Accepted for Publication)

    2018

  11. [19]

    Hansen, N. R. et al. (2003) Geometric Ergodicity of Discrete-time Approximations to Multivariate Diffusions . Bernoulli, 9, 725--743

    2003

  12. [20]

    (2023) Drawing exact samples: rejection sampling, density fusion and constrained disaggregation

    Hu, S. (2023) Drawing exact samples: rejection sampling, density fusion and constrained disaggregation. Ph.D. thesis, University of Essex

    2023

  13. [21]

    and Stoltz, G

    Leli \`e vre, T., Rousset, M. and Stoltz, G. (2019) Hybrid Monte Carlo Methods for Sampling Probability Measures on Submanifolds . Numerische Mathematik, 143, 379--421

    2019

  14. [22]

    and Fang, F

    Li, S., Yang, W., Huang, S., Chen, R., Cheng, X., Zhou, S., Gong, J., Qian, H. and Fang, F. (2023) A hierarchical constraint-based graph neural network for imputing urban area data. International Journal of Geographical Information Science, 37, 1998--2019

    2023

  15. [23]

    Meng, F., Zeng, X.-J., Zhang, Y., Dent, C. J. and Gong, D. (2018) An Integrated Optimization+ Learning Approach to Optimal Dynamic Pricing for the Retailer with Multi-type Customers in Smart Grids . Information Sciences, 448, 215--232

    2018

  16. [24]

    Nicolaescu, L. I. (2020) Lectures on the Geometry of Manifolds. World Scientific

    2020

  17. [25]

    and Matthew, R

    Peppanen, J., Zhang, X., Grijalva, S. and Matthew, R. (2016) Handling Bad or Missing Smart Meter Data through Advanced Data Imputation . Conference: IEEE PES Innovative Smart Grid Technologies (ISGT), September, DOI: 10.1109/ISGT.2016.7781213

    2016

  18. [26]

    and Wang, L

    P \"o tzelberger, K. and Wang, L. (2001) Boundary crossing probability for brownian motion. Journal of applied probability, 38, 152--164

    2001

  19. [27]

    and Mark, M

    Poursharif, G., Brint, A., Black, M. and Mark, M. (2017) Analysing the Ability of Smart Meter Data Toprovide Accurate Information to the UKDNOs . 24th International Conference & Exhibition on Electricity Distribution (CIRED), September

    2017

  20. [28]

    R Foundation for Statistical Computing, Vienna, Austria

    R Core Team (2024) R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. ://www.R-project.org/

    2024

  21. [29]

    N., Moayedi, S., Ahn, C

    Rafsanjani, H. N., Moayedi, S., Ahn, C. R. and Alahmad, M. (2020) A Load-disaggregation Framework to Sense Personalized Energy-use Information in Commercial Buildings . Energy and Buildings, 207, 109633

    2020

  22. [30]

    and Lindsey, J

    Swihart, B. and Lindsey, J. (2022) rmutil: Utilities for Nonlinear Regression and Repeated Measurements Models. ://CRAN.R-project.org/package=rmutil. R package version 1.1.10

    2022

  23. [31]

    Tu, L. W. (2011) An Introduction to Manifolds . Springer

    2011

  24. [32]

    (2018) Sampling the Multivariate Standard Normal Distribution Under a Weighted sum Constraint

    Vrins, F. (2018) Sampling the Multivariate Standard Normal Distribution Under a Weighted sum Constraint . Risks, 6, 64

    2018

  25. [33]

    and Li, F

    Wang, S., Li, R., Evans, A. and Li, F. (2020) Regional Nonintrusive Load Monitoring for Low Voltage Substations and Distributed Energy Resources . Applied Energy, 260, 114225

    2020

  26. [34]

    and Goodman, J

    Zappa, E., Holmes-Cerfon, M. and Goodman, J. (2018) Monte Carlo on Manifolds: Sampling Densities and Integrating Functions . Communications on Pure and Applied Mathematics, 71, 2609--2647

    2018

  27. [35]

    and Stankovic, V

    Zhao, B., Ye, M., Stankovic, L. and Stankovic, V. (2020) Non-intrusive Load Disaggregation Solutions For Very Low-rate Smart Meter Data . Applied Energy, 268, 114949

    2020

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.