Pith. sign in

REVIEW 4 major objections 6 minor 60 references

Estimation of System Parameters Including Repeated Cross-Sectional Data through Emulator-Informed Deep Generative Model

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A physics-informed hypernetwork emulator wired into a Wasserstein GAN recovers parameter distributions of differential equations from unpaired repeated cross-sectional snapshots, including multimodal cases.

desk verdict A sensible new architecture for estimating parameter distributions from repeated cross-sectional data, but the identifiability problem it concedes means the strong performance claims should be taken as empirical demonstrations, not general guarantees. read the letter →

arxiv 2412.19517 v1 pith:2CRXAWPC submitted 2024-12-27 cs.LG cs.AIcs.NAmath.NAq-bio.PEstat.ML

Hyunwoo Cho , Sung Woong Cho , Hyeontae Jo , Hyung Ju Hwang This is my paper · ORCID
classification cs.LGcs.AIcs.NAmath.NAq-bio.PEstat.ML
keywords repeatedcross-sectionaldataparameterestimationdifferentialequationsphysics-informedneuralnetworkshypernetworksWassersteingenerativeadversarialLorenzsystemamyloidbeta
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 a deep generative pipeline can recover the distribution of unknown parameters in an ordinary differential equation when the available data are repeated cross-sectional (RCS) snapshots, i.e., measurements taken at different times from different subjects with no trajectories linking them. The proposed method, EIDGM, wires a physics-informed hypernetwork emulator (HyperPINN) into a Wasserstein GAN: the generator proposes parameter vectors, the emulator instantly produces the corresponding ODE solutions, and the discriminator compares the resulting snapshots with the RCS data. The authors claim that minimizing this distributional mismatch makes EIDGM accurately capture true parameter distributions, including bimodal and trimodal shapes, for exponential growth, logistic growth, and the Lorenz system. A real-data application to amyloid-beta biomarker measurements suggests the method can separate subpopulations with different growth rates and carrying capacities. If true, this gives a practical route to parameter inference and heterogeneity discovery in fields where only unpaired snapshots are available.

What carries the argument

The load-bearing machinery is the HyperPINN emulator coupled with a Wasserstein GAN with gradient penalty. HyperPINN is a physics-informed hypernetwork: a hypernetwork h maps the parameter vector p to the weights θ_m of a main network, so the main network outputs y(t;p) instantly for any p after training on a data loss plus a physics residual loss. The WGAN's generator maps latent noise to candidate parameters; these pass through the emulator to produce synthetic RCS snapshots; the discriminator approximates the Wasserstein distance between synthetic and real snapshots under Kantorovich-Rubinstein duality. Training alternates between discriminator and generator, and the final generator is the estimated parameter distribution.

What would settle it

Build two distinct parameter distributions for one of the paper's models (e.g., a unimodal and a bimodal distribution of the logistic growth rate r) whose emulator-generated RCS snapshots have the same distribution at the five observation times, then feed data from one distribution into EIDGM. If the recovered distribution is the other one, the claim that EIDGM accurately captures true parameter distributions fails. A simpler check is to estimate the distribution twice from different random seeds and compare the Wasserstein distance between the two estimates with the reported distance to the truth.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that matching the distribution of emulator-generated trajectories to the distribution of RCS snapshots is enough to recover the true parameter distribution of the underlying ODE. By updating the WGAN generator and discriminator while passing generated parameters through HyperPINN, EIDGM “accurately captures true parameter distributions” (Section III.A). This is demonstrated on exponential growth, logistic population, and Lorenz-system models with unimodal, bimodal, and trimodal parameter distributions; in the reported comparisons EIDGM achieves the lowest summed one-dimensional Wasserstein distances in most setups, and unlike GP-based calibration it preserves multimodality. Applied to real amyloid-beta 40/42 datasets, the method finds at least two growth-rate patterns in both biomarkers and reveals that carrying capacity is consistent for amyloid-beta 40 but individual-dependent for amyloid-beta 42.

Load-bearing premise

The assumption that carries the whole argument is that matching the distribution of emulator-generated snapshots to the observed RCS snapshot distribution identifies the true parameter distribution; the paper itself notes that generated trajectories can penetrate the RCS data, so this map may not be one-to-one.

Editorial extensions

If this is right

  • For RCS data, EIDGM can replace mean-trajectory or GP-based calibration, preserving information that those methods discard and recovering multimodal parameter distributions.
  • The method can be applied to any ODE for which a fast differentiable emulator can be trained, including chaotic systems like the Lorenz system, though the emulator choice affects accuracy.
  • In the amyloid-beta application, EIDGM separates subjects into at least two growth-rate groups and distinguishes consistent versus individual-dependent carrying capacity, which can inform disease-progression modeling.
  • The comparison with DeepONet+WGAN shows the emulator is a tunable component: for the tri-modal exponential case DeepONet outperformed HyperPINN, so the best emulator depends on the equation.
  • Because only snapshots are needed, the same pipeline could be used in economics and political science on survey or polling data where individuals are not tracked over time.

Reading between the lines

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

  • A likely hidden cost is identifiability: if two parameter distributions push forward to the same snapshot distribution through the emulator, the WGAN has no incentive to distinguish them, so the method should be paired with a pushforward-matching diagnostic or prior constraints.
  • The full-batch WGAN design limits scalability; for large RCS datasets, using minibatches with some variance reduction or an amortized critic would be a natural extension.
  • The same architecture could be adapted to infer distributions of initial conditions or forcing terms, not just parameters, by changing what the generator outputs.
  • If the emulator is made differentiable in the parameters, one could add a regularization term that penalizes implausible parameter regions, which might cure part of the trajectory-penetration issue noted in Section III.D.
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

4 major / 6 minor

Summary. The manuscript proposes EIDGM, a two-stage deep generative method for estimating the distribution of ODE parameters from repeated cross-sectional (RCS) data. A HyperPINN is trained offline as an emulator that maps parameters to trajectory snapshots; a WGAN with gradient penalty is then trained so that generated (t, y) points mimic the RCS dataset. The method is tested on exponential growth, logistic growth, and the Lorenz system under unimodal, bimodal, and trimodal parameter distributions, and is compared with GP-based calibration and DeepONet+WGAN. A real-data application estimates logistic growth parameters for amyloid-beta 40/42 measurements. The appendix contains convergence and approximation bounds for the HyperPINN emulator.

Significance. If the central claim were fully established, EIDGM would be a practically useful tool for parameter-distribution estimation from cross-sectional snapshots, with applications in biology, economics, and political science. The synthetic benchmarks are genuine: they use externally generated ground-truth parameter distributions, so the main empirical comparison is not circular, and EIDGM achieves lower summed 1D Wasserstein distances than the two baselines in seven of nine scenarios. The theoretical appendix provides a nontrivial error analysis for the HyperPINN component. However, the manuscript does not establish identifiability of the underlying parameter distribution from the WGAN objective, and the quantitative evaluation is weakened by single-run comparisons without noise. These gaps substantially limit the strength of the 'accurately captures true parameter distributions' claim and require revision.

major comments (4)
  1. [II.C, Eq. (8), III.D] The objective in Eq. (8) minimizes the Wasserstein distance between distributions of individual (t, y) points. The RCS data are independent snapshots, so this objective constrains only per-time marginal distributions and leaves cross-time couplings unrestricted; the paper itself notes in Section III.D that generated trajectories 'can penetrate the RCS dataset, suggesting potential identifiability issues in parameter estimation.' Since injectivity of the map from parameter distributions to per-time marginals is neither proven nor tested for the logistic and Lorenz systems, the claim in Section III.A that EIDGM 'accurately captures true parameter distributions' is not established by the given objective. The authors should either prove or empirically test identifiability for the reported models, or explicitly reframe the claim as recovering a distribution consistent with the RCS marginals.
  2. [Appendix A] Theorems A.1-A.3, Lemma 1, Lemma 2, and Proposition A.5 concern only the approximation accuracy of HyperPINN as a DE emulator. They do not analyze the WGAN training objective, the convergence of the generator to a target pi(p), or the equivalence class of parameter distributions that match the RCS marginals. Thus the theoretical appendix does not support the paper's central inference claim; an identifiability or consistency analysis for the full EIDGM pipeline is missing.
  3. [Table I; Sections III.B-D] The quantitative evaluation reports a single run per scenario with no error bars, no random seeds, and no noise in the synthetic data. For a stochastic WGAN pipeline, seven-of-nine wins over baselines cannot be distinguished from seeded variability. Please provide multiple independent runs (at least 5-10 seeds) with standard errors, and add experiments with observational noise to show the method is not relying on exact deterministic snapshots. In addition, the summed 1D Wasserstein distance does not detect mismatches in the joint parameter distribution; a joint metric (e.g., energy distance or multivariate Wasserstein) should be reported.
  4. [IV.A] The real-data validation draws 1,000 trajectories from the estimated parameters and compares them with the same RCS data used for fitting. This is an in-sample fit check: it confirms the generator can reproduce the observed marginals, but it does not independently validate the estimated parameter distribution. A hold-out split or a cross-validation scheme (e.g., train on a subset of time points/subjects and predict the rest) would be needed to support the claim that the estimated distributions are 'sufficiently close to the underlying parameter distributions.'
minor comments (6)
  1. [II.C] The generator loss and the alternating update rule are not explicitly specified; only the discriminator loss LD is given. Please state the generator objective (e.g., -1/N sum D(tilde_t_k, tilde_y_k; theta_D)) and the update schedule for reproducibility.
  2. [II.C, Eq. (8)] The definition of Lip(R^d, R) is written as all functions with finite Lipschitz seminorm, while the constraint in Eq. (8) requires ||f||_Lip <= 1. Please reconcile the notation so the '1-Lipschitz' condition is stated unambiguously.
  3. [II.B, Appendix A] The main-text discretized losses in Eqs. (6)-(7) are unnormalized sums, whereas the appendix defines L_physics^(disc) with 1/T_col and 1/N_p averaging. This inconsistency should be fixed so the theory matches the implemented loss.
  4. [III.A] The text says 'four different time-evolutionary DEs' but lists three models; please correct the count.
  5. [II.B/II.C] The symbol D is used both for the parameter sampling distribution in the emulator and for the discriminator; please use distinct notation to avoid confusion.
  6. [Fig. 4 caption] The caption refers to 'six parameters included in the model' although the Lorenz system has three parameters; please correct the caption or clarify what six quantities are meant.

Circularity Check

1 steps flagged · score 2.0 of 10

No load-bearing circularity; the only mild circular step is an auxiliary in-sample real-data validation, which the paper itself flags as an identifiability limitation.

  1. fitted input called prediction [Section IV.A (real-world RCS data application), Fig. 5 caption]
    "We first validate the accuracy of parameter estimates. Unlike the simulation dataset, we cannot find the underlying distributions. Hence, we directly draw 1,000 solution trajectories of the logistic model with the parameter estimates (Fig. 5(a-b), left black lines). This shows that the model solutions with estimated parameters closely match the given RCS data. Therefore, we expect that the estimated distributions are sufficiently close to the underlying parameter distributions."

    The parameter estimates were obtained by minimizing the WGAN discrepancy in Eq. (8) against the same RCS dataset Y. Drawing 1,000 trajectories from those estimates and confirming that they match Y therefore checks the fitting target rather than providing independent evidence for the recovered parameter distribution. The inference that the estimated distribution is close to the true underlying distribution rests on this in-sample agreement, which any distribution whose per-time marginals match the observed RCS snapshots could also satisfy. The paper itself concedes in Section III.D that generated trajectories can 'penetrate the RCS dataset, suggesting potential identifiability issues,' so this validation cannot by itself certify the recovered parameter distribution.

full rationale

The central claim that EIDGM 'accurately captures true parameter distributions' is supported by externally specified synthetic benchmarks (Table I; Figs. 2-4), not by the training objective alone, so it is not circular. The appendix proves only that HyperPINN can approximate DE solutions (Theorems A.1-A.3, Proposition A.5); it does not prove identifiability of the parameter distribution, and the paper explicitly lists the identifiability problem as a limitation in the Conclusion and in Section III.D. The real-data 'validation' in Section IV.A is in-sample, but the paper marks it as an expectation rather than an independent confirmation, and it is not the basis of the synthetic empirical claims. Self-citations to [14], [37], [52], and [53] motivate design choices but are not load-bearing: the emulator is benchmarked against GP and DeepONet, and the approximation argument also relies on external sources ([24], [55], [56], [57]). Overall, no derivation step reduces to its own input by construction; the only mild circularity is the auxiliary real-data consistency check, which does not undermine the externally benchmarked central result.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the empirical WGAN parameter-recovery step, which is not theoretically analyzed; the appendix only shows HyperPINN can approximate ODE solutions. Synthetic benchmark distributions and all network hyperparameters are chosen by hand, and no noise is added, so the evaluation relies on those choices.

free parameters (6)
  • Loss weights alpha and beta = alpha=1; beta=1e-2 for Exponential/Logistic, beta=0 for Lorenz.
    Chosen by hand to balance data and physics loss; no sensitivity analysis is reported.
  • Emulator training sample count Np = 102 (Exponential), 2e2 (Logistic), 1e3 (Lorenz).
    Hand-selected per experiment; this is the sample size in the discretized physics loss and in Theorem A.3.
  • Network widths and depths = Widths 32-128, depths 3-4, tanh activations (Table V).
    Architecture hyperparameters for HyperPINN, generator, and discriminator are chosen by hand without ablation.
  • WGAN training epochs = 5e4 to 1e5 depending on model.
    Training length is selected per experiment; no convergence criterion or early stopping is described.
  • Generator noise dimension = 16 (Exponential/Logistic) or 32 (Lorenz).
    Latent dimension of the WGAN generator is chosen by hand.
  • Synthetic peak locations and ranges = Tables II-IV; hand-selected ranges and peak positions for each experiment.
    The evaluation distributions are constructed by the authors; they are not fitted to data but define what success means.
assumptions (4)
  • standard math f in the ODE is Lipschitz continuous in y and the initial value problem has a unique solution.
    Used by Theorem A.1 and Lemma 1 to bound emulator error and loss Lipschitz constants.
  • domain assumption Parameter distributions D share a common compact support and all network weights are bounded in [-R,R].
    Assumed in Appendix A so the covering number C(epsilon/16, M) is finite (Lemma 2); the paper calls this practically reasonable for real-world bounded parameter ranges.
  • standard math Universal approximation holds for 2-layer fully connected neural networks with C1 activation.
    Used in Proposition A.5 through Theorem A.4 to show the discretized physics loss can be made arbitrarily small.
  • domain assumption Minimizing the Wasserstein distance between emulator-generated trajectories and RCS data drives the generator distribution toward the true parameter distribution.
    This is the load-bearing step of EIDGM in Section II.C; the paper never proves injectivity or convergence of this map and even flags identifiability issues in Section III.D.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Estimation of System Parameters Including Repeated Cross-Sectional Data through Emulator-Informed Deep Generative Model." pith.science (2026). https://pith.science/paper/2CRXAWPC

@misc{pith2026241219517,
  author       = {Pith},
  title        = {Pith review of: Estimation of System Parameters Including Repeated Cross-Sectional Data through Emulator-Informed Deep Generative Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2CRXAWPC}},
  note         = {Machine review of arXiv:2412.19517}
}
read the original abstract

Differential equations (DEs) are crucial for modeling the evolution of natural or engineered systems. Traditionally, the parameters in DEs are adjusted to fit data from system observations. However, in fields such as politics, economics, and biology, available data are often independently collected at distinct time points from different subjects (i.e., repeated cross-sectional (RCS) data). Conventional optimization techniques struggle to accurately estimate DE parameters when RCS data exhibit various heterogeneities, leading to a significant loss of information. To address this issue, we propose a new estimation method called the emulator-informed deep-generative model (EIDGM), designed to handle RCS data. Specifically, EIDGM integrates a physics-informed neural network-based emulator that immediately generates DE solutions and a Wasserstein generative adversarial network-based parameter generator that can effectively mimic the RCS data. We evaluated EIDGM on exponential growth, logistic population models, and the Lorenz system, demonstrating its superior ability to accurately capture parameter distributions. Additionally, we applied EIDGM to an experimental dataset of Amyloid beta 40 and beta 42, successfully capturing diverse parameter distribution shapes. This shows that EIDGM can be applied to model a wide range of systems and extended to uncover the operating principles of systems based on limited data.

Figures

Figures reproduced from arXiv: 2412.19517 by the authors.

Figure 1
Figure 1. Schematic of Emulator-Informed Deep-Generative Model (EIDGM) A generator in WGAN is trained to produce a possible set of parameters. These parameters are used as inputs of HyperPINN, immediately producing solutions corresponding to the set of parameters. We then calculate the loss through the discriminator in WGAN by measuring the difference (distribution) between solutions and RCS data. By minimizing the loss, the … view at source ↗
Figure 2
Figure 2. Visualization of parameter estimates for exponential growth model with three different emulators: Gaussian Process (GP), DeepONet (with WGAN), and EIDGM (HyperPINN with WGAN).(a) If data are obtained from an exponential model with one parameter peak (RCS Data), all methods accurately estimate the underlying parameters (Right, Estimation). (b), (c) When data are obtained with multiple parameter peaks, GP fails to est… view at source ↗
Figure 3
Figure 3. Visualization of estimation results for logistic population model with three different emulators: GP, DeepONet, and HyperPINN+WGAN (EIDGM).(a) If data are obtained from a logistic model with one parameter peak (Left, Observation), all methods accurately estimate the underlying parameters (Right, Estimation). (b) (c) Only EIDGM accurately estimates the parameters when there are two or three parameter peaks. on temper… view at source ↗
Figures from the paper (1 more)
Figure 5
Figure 5. Figure 5: Estimation results for amyloid beta accumulation in real exper￾imental datasets using a logistic model with EIDGM. (a-b) We utilized EIDGM with two distinct datasets (left): Aβ40 and Aβ42 (left red dots). The parameter estimates revealed different patterns (right black…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 48 canonical work pages

  1. [1]

    Density physics-informed neural networks reveal sources of cell heterogeneity in signal transduction,

    H. Jo, H. Hong, H. J. Hwang, W. Chang, and J. K. Kim, “Density physics-informed neural networks reveal sources of cell heterogeneity in signal transduction,” Patterns, vol. 5, no. 2, 2024

    work page 2024

  2. [2]

    Random coefficient models for time-series—cross-section data: Monte carlo experi- ments,

    N. Beck and J. N. Katz, “Random coefficient models for time-series—cross-section data: Monte carlo experi- ments,” Political Analysis, vol. 15, no. 2, pp. 182–195, 2007

    work page 2007

  3. [3]

    Modeling dynamics in time-series–cross-section political economy data,

    ——, “Modeling dynamics in time-series–cross-section political economy data,” Annual review of political sci- ence, vol. 14, no. 1, pp. 331–352, 2011

    work page 2011

  4. [4]

    Pan, Repeated Cross-Sectional Design

    X. Pan, Repeated Cross-Sectional Design . Cham: Springer International Publishing, 2019, pp. 1–5

    work page 2019

  5. [5]

    G. W. J. M. Stevens, S. Van Dorsselaer, M. Boer, S. De Roos, E. Duinhof, T. Ter Bogt, R. Van Den Eijn- den, L. Kuyper, D. Visser, W. A. M. V ollebergh et al. , HBSC 2017. Gezondheid en welzijn van jongeren in Nederland. Utrecht, Netherlands: Utrecht University, 2018

    work page 2017

  6. [6]

    Bryman, Social research methods

    A. Bryman, Social research methods . Oxford, UK: Oxford university press, 2016

    work page 2016

  7. [7]

    Systematic modeling-driven experiments identify distinct molecular clockworks un- derlying hierarchically organized pacemaker neurons,

    E. M. Jeong, M. Kwon, E. Cho, S. H. Lee, H. Kim, E. Y . Kim, and J. K. Kim, “Systematic modeling-driven experiments identify distinct molecular clockworks un- derlying hierarchically organized pacemaker neurons,” Proceedings of the National Academy of Sciences , vol. 119, no. 8, p. e2113403119, 2022

    work page 2022

  8. [8]

    The radiosensitizer onalespib increases complete remis- sion in 177 lu-dotatate-treated mice bearing neuroen- docrine tumor xenografts,

    S. Lundsten, D. Spiegelberg, N. R. Raval, and M. Nestor, “The radiosensitizer onalespib increases complete remis- sion in 177 lu-dotatate-treated mice bearing neuroen- docrine tumor xenografts,” European journal of nuclear medicine and molecular imaging , vol. 47, pp. 980–990, 2020

    work page 2020

Show all 60 references
  1. [9]

    Men, women and the dynamics of presidential approval,

    H. D. Clarke, M. C. Stewart, M. Ault, and E. El- liott, “Men, women and the dynamics of presidential approval,” British Journal of Political Science , vol. 35, no. 1, pp. 31–51, 2005

    2005

  2. [10]

    Whose economy? perceptions of national economic performance during unequal growth,

    D. J. Hopkins, “Whose economy? perceptions of national economic performance during unequal growth,” Public Opinion Quarterly, vol. 76, no. 1, pp. 50–71, 2012

    2012

  3. [11]

    Strategic party government: Party influence in congress, 1789– 2000,

    M. J. Lebo, A. J. McGlynn, and G. Koger, “Strategic party government: Party influence in congress, 1789– 2000,” American Journal of Political Science , vol. 51, no. 3, pp. 464–481, 2007

    2000

  4. [12]

    J. A. Segal and H. J. Spaeth, The Supreme Court and the attitudinal model revisited. Cambridge, UK: Cambridge University Press, 2002

    2002

  5. [13]

    B. D. Wood, The myth of presidential representation . Cambridge, UK: Cambridge University Press, 2009

    2009

  6. [14]

    Estimating the distribution of parameters in differential equations with repeated cross-sectional data,

    H. Jo, S. W. Cho, and H. J. Hwang, “Estimating the distribution of parameters in differential equations with repeated cross-sectional data,” PLOS Computational Bi- ology, vol. 20, no. 12, p. e1012696, 2024. 8

    2024

  7. [15]

    Host-pathogen kinetics during influenza infection and coinfection: insights from predictive mod- eling,

    A. M. Smith, “Host-pathogen kinetics during influenza infection and coinfection: insights from predictive mod- eling,” Immunological reviews, vol. 285, no. 1, pp. 97– 112, 2018

    2018

  8. [16]

    Parameter and uncertainty estimation for dynamical systems using surrogate stochastic processes,

    M. Chung, M. Binois, R. B. Gramacy, J. M. Bardsley, D. J. Moquin, A. P. Smith, and A. M. Smith, “Parameter and uncertainty estimation for dynamical systems using surrogate stochastic processes,” SIAM Journal on Scien- tific Computing, vol. 41, no. 4, pp. A2212–A2238, 2019

    2019

  9. [17]

    Improved most likely het- eroscedastic gaussian process regression via bayesian residual moment estimator,

    Q.-H. Zhang and Y .-Q. Ni, “Improved most likely het- eroscedastic gaussian process regression via bayesian residual moment estimator,”IEEE Transactions on Signal Processing, vol. 68, pp. 3450–3460, 2020

    2020

  10. [18]

    hetgp: Heteroskedastic gaussian process modeling and sequential design in r,

    M. Binois and R. B. Gramacy, “hetgp: Heteroskedastic gaussian process modeling and sequential design in r,” Journal of Statistical Software , vol. 98, pp. 1–44, 2021

    2021

  11. [19]

    Markov chain monte carlo without likelihoods,

    P. Marjoram, J. Molitor, V . Plagnol, and S. Tavar ´e, “Markov chain monte carlo without likelihoods,” Pro- ceedings of the National Academy of Sciences , vol. 100, no. 26, pp. 15 324–15 328, 2003

    2003

  12. [20]

    The frontier of simulation-based inference,

    K. Cranmer, J. Brehmer, and G. Louppe, “The frontier of simulation-based inference,” Proceedings of the National Academy of Sciences , vol. 117, no. 48, pp. 30 055– 30 062, 2020

    2020

  13. [21]

    Equation of state calculations by fast computing machines,

    N. Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller, and E. Teller, “Equation of state calculations by fast computing machines,” The journal of chemical physics, vol. 21, no. 6, pp. 1087–1092, 1953

    1953

  14. [22]

    Monte carlo sampling methods using markov chains and their applications,

    W. K. Hastings, “Monte carlo sampling methods using markov chains and their applications,” Biometrika, 1970

    1970

  15. [23]

    On the importance of the jacobian determinant in parameter inference for random parameter and random measurement error models,

    D. Swigon, S. R. Stanhope, S. Zenker, and J. E. Ru- bin, “On the importance of the jacobian determinant in parameter inference for random parameter and random measurement error models,” SIAM/ASA Journal on Un- certainty Quantification , vol. 7, no. 3, pp. 975–1006, 2019

    2019

  16. [24]

    Hyper- pinn: Learning parameterized differential equations with physics-informed hypernetworks,

    F. de Avila Belbute-Peres, Y .-f. Chen, and F. Sha, “Hyper- pinn: Learning parameterized differential equations with physics-informed hypernetworks,” in The symbiosis of deep learning and differential equations , 2021

    2021

  17. [25]

    Wasserstein generative adversarial networks,

    M. Arjovsky, S. Chintala, and L. Bottou, “Wasserstein generative adversarial networks,” in International confer- ence on machine learning . PMLR, 2017, pp. 214–223

    2017

  18. [26]

    Few-shot prediction of amyloid β accumulation from mainly unpaired data on biomarker candidates,

    Y . Yada and H. Naoki, “Few-shot prediction of amyloid β accumulation from mainly unpaired data on biomarker candidates,” NPJ Systems Biology and Applications , vol. 9, no. 1, p. 59, 2023

    2023

  19. [27]

    The deep ritz method: a deep learning- based numerical algorithm for solving variational prob- lems,

    B. Yu et al. , “The deep ritz method: a deep learning- based numerical algorithm for solving variational prob- lems,” Communications in Mathematics and Statistics , vol. 6, no. 1, pp. 1–12, 2018

    2018

  20. [28]

    Physics- informed neural networks: A deep learning framework for solving forward and inverse problems involving non- linear partial differential equations,

    M. Raissi, P. Perdikaris, and G. E. Karniadakis, “Physics- informed neural networks: A deep learning framework for solving forward and inverse problems involving non- linear partial differential equations,” Journal of Compu- tational physics, vol. 378, pp. 686–707, 2019

    2019

  21. [29]

    Deepsdf: Learning continuous signed dis- tance functions for shape representation,

    J. J. Park, P. Florence, J. Straub, R. Newcombe, and S. Lovegrove, “Deepsdf: Learning continuous signed dis- tance functions for shape representation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 165–174

    2019

  22. [30]

    Learning implicit fields for generative shape modeling,

    Z. Chen and H. Zhang, “Learning implicit fields for generative shape modeling,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 5939–5948

    2019

  23. [31]

    Occupancy networks: Learning 3d re- construction in function space,

    L. Mescheder, M. Oechsle, M. Niemeyer, S. Nowozin, and A. Geiger, “Occupancy networks: Learning 3d re- construction in function space,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 4460–4470

    2019

  24. [32]

    Learning nonlinear operators via deeponet based on the universal approximation theorem of operators,

    L. Lu, P. Jin, G. Pang, Z. Zhang, and G. E. Karniadakis, “Learning nonlinear operators via deeponet based on the universal approximation theorem of operators,” Nature machine intelligence, vol. 3, no. 3, pp. 218–229, 2021

    2021

  25. [33]

    Shift-deeponet: Extending deep operator networks for discontinuous output functions,

    P. S. Hadorn, “Shift-deeponet: Extending deep operator networks for discontinuous output functions,” Master’s Thesis, ETH Zurich, Seminar for Applied Mathematics, 2022. [Online]. Available: https://www. research-collection.ethz.ch/handle/20.500.11850/539793

    2022

  26. [34]

    NOMAD: Nonlinear manifold decoders for operator learning,

    J. H. Seidman, G. Kissas, P. Perdikaris, and G. J. Pappas, “NOMAD: Nonlinear manifold decoders for operator learning,” in Advances in Neural Information Processing Systems, A. H. Oh, A. Agarwal, D. Belgrave, and K. Cho, Eds., 2022. [Online]. Available: https: //openreview.net/...

    2022

  27. [35]

    Hypernetworks,

    D. Ha, A. Dai, and Q. V . Le, “Hypernetworks,” arXiv preprint arXiv:1609.09106, 2016

    2016 arXiv

  28. [36]

    On the modularity of hyper- networks,

    T. Galanti and L. Wolf, “On the modularity of hyper- networks,” Advances in Neural Information Processing Systems, vol. 33, pp. 10 409–10 419, 2020

    2020

  29. [37]

    Hyperdeeponet: learning operator with complex target function space using the limited resources via hypernetwork,

    J. Y . Lee, S. W. Cho, and H. J. Hwang, “Hyperdeeponet: learning operator with complex target function space using the limited resources via hypernetwork,” arXiv preprint arXiv:2312.15949, 2023

    2023 arXiv

  30. [38]

    Generative adversarial nets,

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio, “Generative adversarial nets,” Advances in neural infor- mation processing systems , vol. 27, 2014

    2014

  31. [39]

    A framework for data-driven solution and parameter estimation of pdes using conditional generative adversarial networks,

    T. Kadeethum, D. O’Malley, J. N. Fuhg, Y . Choi, J. Lee, H. S. Viswanathan, and N. Bouklas, “A framework for data-driven solution and parameter estimation of pdes using conditional generative adversarial networks,” Na- ture Computational Science, vol. 1, no. 12, pp. 819–829, 2021

    2021

  32. [40]

    Solution of physics-based bayesian inverse problems with deep gen- erative priors,

    D. V . Patel, D. Ray, and A. A. Oberai, “Solution of physics-based bayesian inverse problems with deep gen- erative priors,” Computer Methods in Applied Mechanics and Engineering, vol. 400, p. 115428, 2022

    2022

  33. [41]

    Conditional generative ad- versarial nets,

    M. Mirza and S. Osindero, “Conditional generative ad- versarial nets,” arXiv preprint arXiv:1411.1784 , 2014

    2014 arXiv

  34. [42]

    Gatsbi: Generative adversarial train- ing for simulation-based inference,

    P. Ramesh, J.-M. Lueckmann, J. Boelts, ´A. Tejero- Cantero, D. S. Greenberg, P. J. Gonc ¸alves, and J. H. Macke, “Gatsbi: Generative adversarial train- ing for simulation-based inference,” arXiv preprint arXiv:2203.06481, 2022

    2022 arXiv

  35. [43]

    Attention is all you need,

    A. Vaswani, “Attention is all you need,” arXiv preprint 9 arXiv:1706.03762, 2017

    2017 arXiv

  36. [44]

    Denoising diffu- sion implicit models,

    J. Song, C. Meng, and S. Ermon, “Denoising diffu- sion implicit models,” arXiv preprint arXiv:2010.02502 , 2020

    2010 arXiv

  37. [45]

    All-in-one simulation-based inference,

    M. Gloeckler, M. Deistler, C. Weilbach, F. Wood, and J. H. Macke, “All-in-one simulation-based inference,” arXiv preprint arXiv:2404.09636 , 2024

    2024 arXiv

  38. [46]

    Improved training of wasserstein gans,

    I. Gulrajani, F. Ahmed, M. Arjovsky, V . Dumoulin, and A. C. Courville, “Improved training of wasserstein gans,” Advances in neural information processing sys- tems, vol. 30, 2017

    2017

  39. [47]

    Villani et al., Optimal transport: old and new

    C. Villani et al., Optimal transport: old and new. Berlin, Germany: Springer, 2009, vol. 338

    2009

  40. [48]

    Deeponet: Learning nonlinear operators for identifying differential equations based on the universal approximation theorem of opera- tors,

    L. Lu, P. Jin, and G. E. Karniadakis, “Deeponet: Learning nonlinear operators for identifying differential equations based on the universal approximation theorem of opera- tors,” arXiv preprint arXiv:1910.03193 , 2019

    1910 arXiv

  41. [49]

    Spatiotem- poral distribution of β-amyloid in alzheimer disease is the result of heterogeneous regional carrying capacities,

    A. Whittington, D. J. Sharp, and R. N. Gunn, “Spatiotem- poral distribution of β-amyloid in alzheimer disease is the result of heterogeneous regional carrying capacities,” Journal of Nuclear Medicine, vol. 59, no. 5, pp. 822–827, 2018

    2018

  42. [50]

    Optimal anti- amyloid-beta therapy for alzheimer’s disease via a per- sonalized mathematical model,

    W. Hao, S. Lenhart, and J. R. Petrella, “Optimal anti- amyloid-beta therapy for alzheimer’s disease via a per- sonalized mathematical model,” PLoS computational bi- ology, vol. 18, no. 9, p. e1010481, 2022

    2022

  43. [51]

    Dgm: A deep learning algorithm for solving partial differential equations,

    J. Sirignano and K. Spiliopoulos, “Dgm: A deep learning algorithm for solving partial differential equations,” Jour- nal of computational physics , vol. 375, pp. 1339–1364, 2018

    2018

  44. [52]

    Deep neural network approach to forward-inverse problems,

    H. Jo, H. Son, H. J. Hwang, and E. Kim, “Deep neural network approach to forward-inverse problems,” arXiv preprint arXiv:1907.12925, 2019

    1907 arXiv

  45. [53]

    The monotone traveling wave solution of a bistable three-species compe- tition system via unconstrained neural networks,

    S. W. Cho, S. Hwang, and H. J. Hwang, “The monotone traveling wave solution of a bistable three-species compe- tition system via unconstrained neural networks,” Mathe- matical Biosciences and Engineering , vol. 20, no. 4, pp. 7154–7170, 2023

    2023

  46. [54]

    Physics- informed neural operator for learning partial differential equations,

    Z. Li, H. Zheng, N. Kovachki, D. Jin, H. Chen, B. Liu, K. Azizzadenesheli, and A. Anandkumar, “Physics- informed neural operator for learning partial differential equations,” ACM/JMS Journal of Data Science , vol. 1, no. 3, pp. 1–27, 2024

    2024

  47. [55]

    M. W. Hirsch, S. Smale, and R. L. Devaney, Differential equations, dynamical systems, and an introduction to chaos. San Diego, CA: Academic press, 2013

    2013

  48. [56]

    A model of inductive bias learning,

    J. Baxter, “A model of inductive bias learning,” Journal of artificial intelligence research , vol. 12, pp. 149–198, 2000

    2000

  49. [57]

    Simultaneous approximations of multivariate functions and their derivatives by neural networks with one hidden layer,

    X. Li, “Simultaneous approximations of multivariate functions and their derivatives by neural networks with one hidden layer,” Neurocomputing, vol. 12, no. 4, pp. 327–343, 1996. VII. B IOGRAPHY SECTION Hyunwoo Cho received the B.S. degree in mathematics from POSTECH, in 2019, ...

    1996

  50. [58]

    Training HyperPINN by minimizing both data and physics losses: Raissi et al. introduced the original physics loss [28], ℓ(nn(t), θnn), which measures how well an artificial neural network nn(t) with weights and biases θnn satisfies the given dynamical system of Eq. (1): ℓ(nn(t...

  51. [59]

    Thus, we alterna- tively use the discretized version of Lphysics defined in Eq

    Difference of the two physics loss functions: In practice, we cannot calculate the above definite integral and expectation due to the lack of computational resources. Thus, we alterna- tively use the discretized version of Lphysics defined in Eq. (7) as follows: Lphysics(disc)...

  52. [60]

    In this section, we demonstrate that hyperPINN can effectively minimize Lphysics(disc) by applying the universal approximation theorem for neural networks

    Minimizing the discretized loss function using synthetic RCS data: The previous section guarantees that the physics loss can be discretized up to a small error when we train it with a sufficient number of parameters. In this section, we demonstrate that hyperPINN can effective...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.