Pith. sign in

REVIEW 3 major objections 6 minor 55 references

Principled model selection for stochastic dynamics

T0 review · 3 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read PASTIS finds true stochastic models that AIC and BIC miss.

desk verdict A genuinely useful p-controlled information criterion for stochastic model selection, with a real but honestly flagged caveat about correlated libraries and multiplicative noise. read the letter →

arxiv 2501.10339 v3 pith:5SB6GJMH submitted 2025-01-17 cond-mat.soft cond-mat.stat-mechphysics.data-anstat.ML

Andonis Gerardos , Pierre Ronceray This is my paper · ORCID
classification cond-mat.softcond-mat.stat-mechphysics.data-anstat.ML MSC 60G7060H1062M0562F03
keywords modelselectionstochasticdifferentialequationssparseidentificationextremevaluetheoryinformationcriteriamultiplehypothesistestinglikelihoodinferencedynamics
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

Learning stochastic differential equations from trajectory data usually means picking a small set of drift functions from a large library. The paper argues that standard selection tools miss a key source of overfitting: the number of candidate models grows combinatorially with library size, so the best-fitting spurious model will eventually beat the true one by chance alone. Its central proposal, PASTIS, adds a penalty proportional to $n_B \log(n_0/p)$ to the log-likelihood, where $n_0$ is the library size and $p$ is a user-set false-positive rate. The paper shows on synthetic benchmarks that PASTIS recovers the exact generating model more often than AIC, BIC, cross-validation, SINDy, and LASSO, and that the same criterion works for stochastic partial differential equations and under sampling and measurement imperfections.

What carries the argument

The load-bearing object is the PASTIS information criterion $I_{\mathrm{PASTIS}}(B) = I(B) - n_B \log(n_0/p)$, built on the quasi-likelihood information $I(B)$ of the paper's estimation approach. It does its work through two statistical facts: Wilks' theorem, which says the information gain from a single superfluous term is asymptotically $\frac12 \chi^2_1$, and extreme value theory, which says the maximum of many such independent gains concentrates near $\log(n_0)$ and fluctuates with a Gumbel law. The $\log(n_0/p)$ penalty is exactly the threshold that keeps the probability that any spurious term beats the true model at approximately $p$, turning a combinatorial search over $2^{n_0}$ candidate models into a tractable, significance-controlled selection rule.

What would settle it

Run PASTIS with a fixed $p$ on a simulated system whose true drift is known but whose candidate library is deliberately redundant (e.g., monomials up to degree 20 on a trajectory that stays in a narrow range, so many functions are nearly collinear), and count how often the selected model contains at least one term outside the true set in the long-trajectory limit; if the measured false-positive rate is substantially larger than $p$ or than the exact finite-library prediction $g(p)$, the independence assumption underlying the Gumbel penalty is the culprit.

Watch

Extended reading notes

Core claim

The paper's central claim is that the right penalty for sparse model selection in stochastic dynamics is not one unit per parameter, as in Akaike's criterion, but $\log(n_0/p)$ per parameter, where $n_0$ is the size of the candidate library and $p$ is the target probability of accepting a superfluous term. The criterion, called PASTIS, is $I_{\mathrm{PASTIS}}(B) = I(B) - n_B \log(n_0/p)$, with $I(B)$ the quasi-likelihood information gain of basis $B$ relative to a zero-force null model. Because the largest information gain among $n_0-n_*$ superfluous terms grows like $\log(n_0)$ (a Gumbel extreme-value effect), a fixed per-parameter penalty like AIC's lets one spurious term slip in with probability approaching one; the $\log(n_0/p)$ penalty instead caps that probability near $p$. The paper demonstrates on four stochastic benchmark models that PASTIS selects the exact generating model with probability above $1-p$ given enough data, and that it outperforms AIC, BIC, cross-validation, SINDy, and LASSO in exact-recovery speed.

Load-bearing premise

The derivation assumes that the information gain of each irrelevant candidate term is statistically independent of the others and follows the same standard chi-squared shape, so the best-scoring irrelevant term obeys the extreme-value formula; when the candidate functions are strongly correlated with each other along the trajectory, that independence fails and the user-chosen false-positive probability $p$ may no longer equal the actual rate of accepting a spurious term.

Editorial extensions

If this is right

  • With enough data, PASTIS recovers the exact sparse generating model in more than $1-p$ of simulation trials, while AIC and cross-validation keep adding superfluous terms even in the long-time limit.
  • BIC eventually selects the true model asymptotically but converges far more slowly than PASTIS, so PASTIS reaches exact recovery with much shorter trajectories.
  • The criterion extends unchanged in form to stochastic partial differential equations when the library includes finite-difference operators, allowing the same procedure to identify reaction-diffusion and other field equations.
  • With modest modifications (a trapezoid likelihood, a three-point diffusion estimator, and a Stratonovich form), PASTIS maintains high exact-match accuracy at large sampling intervals and under measurement noise.
  • In misspecified settings where the true force is not in the library, PASTIS selects smaller, more parsimonious approximations than AIC, BIC, CV, or SINDy while retaining competitive prediction error.

Reading between the lines

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

  • Beyond the paper: the $\log(n_0/p)$ penalty is the same multiplicity correction that physicists call the look-elsewhere effect, so PASTIS could be read as a general template for any likelihood-based search over a large structured hypothesis space, not only SDE drift selection.
  • Beyond the paper: the independence assumption behind the Gumbel bound will be violated when candidate basis functions are strongly correlated along the trajectory, as typical polynomial libraries are; the supplementary material shows one such discrepancy for multiplicative-noise systems, and a natural extension is to replace the single penalty $\log(n_0/p)$ by a data-dependent effective-number-of-
  • Beyond the paper: a testable extension suggested by the paper's own framing is to make $p$ adaptive, small early in the search and larger later, which could speed up convergence to the true model at short times without sacrificing the long-time false-positive guarantee.
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 / 6 minor

Summary. The manuscript introduces Parsimonious Stochastic Inference (PASTIS), a model-selection criterion for stochastic differential equations (SDEs) and stochastic partial differential equations (SPDEs). The criterion is IPASTIS(B) = I(B) − nB log(n0/p), where I(B) is a quasi-likelihood-based information measure, n0 is the size of the candidate library, and p is a user-set significance level. The penalty is derived from extreme-value statistics: if the information gains ΔIs for adding each superfluous basis function are independent and asymptotically 1/2 χ²₁, the maximum gain is Gumbel-distributed, leading to the log(n0/p) term. The authors show that AIC and cross-validation overfit in large libraries, and they benchmark PASTIS against AIC, BIC, SINDy, CV, and LASSO on stochastic Lorenz, Ornstein–Uhlenbeck, Lotka–Volterra, and Gray–Scott models, reporting higher exact-match accuracy and competitive prediction error. Extensions to large sampling intervals, measurement noise, and SPDEs are discussed, and code is provided.

Significance. If the claimed calibration holds, PASTIS is a substantial methodological contribution: it gives a principled, interpretable way to control overfitting risk in sparse SDE discovery, with a single user parameter p and an explicit combinatorial penalty. The derivation chain from Eq. (6) through Wilks' theorem to the Gumbel penalty is transparent and internally consistent under the stated assumptions, and the paper provides an exact finite-n0 correction g(p) in the Supplement. The benchmarks are synthetic but cover a useful range of noise regimes, and the comparison methods are tuned with care. The paper also ships reproducible code and presents falsifiable predictions (e.g., the functional form of the penalty and g(p)). However, the practical validity of the central p-as-significance-level claim is not fully established for the benchmark settings, because the derivation assumes independence and exact Wilks behavior that are not guaranteed for correlated polynomial bases or multiplicative noise. The paper's own Supplement Fig. S1f shows a deviation for the multiplicative-noise Lotka–Volterra case.

major comments (3)
  1. [End Matter 2, Eqs. (8)–(9)] The derivation of the Gumbel penalty treats the information gains ΔIs as independent and asymptotically 1/2 χ²₁, with the stated sufficient condition ⟨s(xt)s′(xt)⟩ = 0 for superfluous functions s ≠ s′. For the polynomial bases used in Fig. 3 (monomials and cross terms), this condition is typically violated along a bounded trajectory: x, x², x³ and cross terms are strongly correlated. The paper does not quantify how such correlations affect the realized false-positive rate, even though p is defined as that rate. I ask the authors to report direct false-positive-rate measurements (using the known true models in the OU and Lorenz benchmarks at large τ) as a function of p, for both uncorrelated and correlated bases, and either confirm that p remains a good approximation or replace the claim with an explicit, evidence-based upper bound.
  2. [Eq. (7) and Supplement Fig. S1f] Wilks' theorem is invoked to justify the 1/2 χ²₁ distribution of ΔIs, but Eq. (3) is a quasi-likelihood for multiplicative noise, not the true likelihood (as the main text acknowledges). The Lotka–Volterra benchmark is multiplicative, and Supplement Fig. S1f shows that the theoretical g(p) is inaccurate in this case. Since this system is one of the four central benchmarks and the central claim is that p is the false-positive probability, the paper should either (i) restrict the significance-level interpretation to additive-noise cases, or (ii) provide a direct empirical calibration of p for multiplicative noise and state the regime of validity. Without this, the claim that PASTIS controls overfitting at a chosen significance level is not established for the multiplicative-noise setting.
  3. [End Matter 3 and Fig. 5 / Fig. S1] Part of the empirical support for PASTIS is that the asymptotic exact-match accuracy converges to a value > 1 − p (e.g., Fig. 3). This comparison is partly definitional, because End Matter 3 defines p as the probability of selecting a one-superfluous-term model. The genuinely testable predictions are the functional form of the penalty and the finite-n0 correction g(p); for the additive-noise benchmarks g(p) matches, which is encouraging. However, the paper should separate the definitional statement (error ≈ p) from the testable prediction (the shape and n0-dependence of g(p)) and present direct false-positive rates versus p for correlated bases. The current Fig. S1 panels do not isolate the effect of basis correlations on the calibration.
minor comments (6)
  1. [Main text, 'Failure of AIC'] There is a typo: 'F ailure of AIC' should read 'Failure of AIC'.
  2. [End Matter 4, Eq. (12)] The displayed equation appears garbled in the typeset version (the term '1q 4⟨ ˆD⟩' in the denominator is not readable); please fix the formatting.
  3. [End Matter 4, text after Eq. (14)] The sentence 'Its minimization leads to the result presented in Fig. 4b' should read 'maximization', since IPASTIS is a criterion to be maximized.
  4. [Supplement Appendix B] The text says AIC, BIC, and CV use the same greedy forward-selection algorithm as employed for PASTIS, while the main text describes PASTIS as hill-climbing with both additions and removals; please clarify whether removals are included for the baselines or whether only forward selection is used.
  5. [End Matter 5 (SPDE)] The transferability of the Wilks theorem and extreme-value argument to discretized fields is asserted rather than demonstrated; given the spatial correlations in the basis (e.g., differential operators), a brief argument or a numerical check would strengthen the SPDE extension.
  6. [Discussion] The paper mentions extended BIC (EBIC) as a consistency perspective but does not include it in the benchmarks; a numerical comparison with EBIC would help position PASTIS relative to existing multiplicity-adjusted criteria.

Circularity Check

1 steps flagged · score 1.0 of 10

Mild self-definitional element: the 1-p asymptotic exact-match plateau restates the definition of p; the core benchmark comparisons are otherwise independent.

  1. self definitional [Eq. (9); End Matter 3; Fig. 3a3-d3]
    "Here, the user-chosen parameter p ≪ 1 is a statistical significance threshold for accepting basis functions. It sets the target probability (approximately p) that the criterion would select a model containing one superfluous basis function rather than the true model (End Matter 3). ... We find that, in all cases, with sufficient amounts of data, the exact match accuracy of PASTIS converges to a value > 1 − p (Figs. 3a3-d3)."

    Eq. 9 uses the penalty nB log(n0/p), and End Matter 3 derives P[max_s IPASTIS(B*+s) > IPASTIS(B*)] ≈ p from that penalty. Thus p is defined as the target false-positive probability, and the reported asymptotic exact-match rate > 1-p is the same statement, under the stated chi-square and independence assumptions, rather than an independently predicted outcome. The empirical agreement confirms the derivation's internal consistency but does not by itself validate the method; the validation comes from external comparisons against AIC, BIC, CV, SINDy, and LASSO, and from the parameter-free external extreme-value result. Because p is a user input rather than a fitted constant, this is a minor, non-load-bearing self-definitional element.

full rationale

The derivation of PASTIS is largely self-contained: starting from the SFI quasi-likelihood (Eq. 3), the paper derives the AIC bias correction (Eq. 6), applies Wilks' theorem (Eq. 7), and uses an external extreme-value result (Leadbetter, Ref. [41]) to obtain Eq. 8, from which IPASTIS(B)=I(B)-nB log(n0/p) is constructed in Eq. 9. The parameter p is a user-set significance threshold, not fitted to the benchmarks, and the claimed advantage over AIC, BIC, CV, SINDy, and LASSO rests on independent synthetic-data comparisons with explicit hyperparameter sweeps (Fig. 3 and Fig. S2). The only mildly circular element is the calibration statement that exact match accuracy converges to a value > 1-p: because p is defined in End Matter 3 as the probability of preferring a one-superfluous-term model, the asymptotic plateau is a restatement of p's meaning under the stated distributional assumptions, not an independent empirical discovery. This is not load-bearing for the method's practical value, which is supported by the external comparisons and by the honest discrepancy reported for multiplicative noise in Supplement Fig. S1f, where the theoretical g(p) is less accurate. The possible violation of the independence assumption for correlated polynomial libraries is a robustness caveat, not a circularity.

Assumptions & free parameters 1 free parameters · 6 assumptions · 0 invented entities

The method's free-parameter count is minimal, with one user-set p, but its theoretical guarantee rests on several explicit assumptions: Wilks asymptotics, ideal diffusion estimation, independence of superfluous terms, well-specified model existence, and large-library asymptotics. The independence assumption is the most fragile and is not validated for the correlated bases used in benchmarks.

free parameters (1)
  • p = 0.001 in main benchmarks; explored over 0.001 to 1 in supplement
    The PASTIS penalty is nB log(n0/p), so p directly controls the penalty magnitude and the false-positive rate. It is an input, not optimized against the benchmark outcomes, but it is a tunable parameter of the method.
assumptions (6)
  • standard math Wilks' theorem gives Delta Is ~ 1/2 chi^2_1 for a superfluous term added to the true model (Eq. 7).
    Used to model the information gain of each unnecessary basis function; requires the quasi-likelihood Eq. 3 to be asymptotically a genuine log-likelihood, which holds only in the ideal additive-noise limit.
  • domain assumption The diffusion estimate Dbar equals the true diffusion D, so the error-likelihood relation Eq. 6 holds without correction.
    Stated in End Matter 1; neglected correlations yield higher-order terms, and multiplicative noise or measurement error introduces biases that the robustness modifications only partially correct.
  • domain assumption The superfluous information gains Delta Is are independent across candidate functions, equivalent to orthogonality <s(xt)s'(xt)> = 0.
    Assumed in End Matter 2 to derive both the Gumbel approximation Eq. 8 and the exact CDF Eq. 11; strongly correlated polynomial libraries violate this, so the p calibration is approximate there.
  • domain assumption A true sparse model B* exists and is contained in the library B0 (well-specified setting).
    Used to define exact match accuracy and the p interpretation; the misspecified case is treated separately in Supplement C with parsimony, not exact recovery, as the goal.
  • standard math For SDEs the BIC penalty uses log tau, and O(1) prior and determinant terms are negligible in the long-trajectory limit.
    Derived in Supplement B.1 (Eqs. S2 to S5) to define the BIC benchmark; the result depends on the approximate likelihood curvature and on treating tau as the effective sample size.
  • domain assumption The library is large, n0 >> n*, so p(n0-n*)/n0 is approximately p in End Matter 3.
    Used to equate 1 - exp(-p(n0-n*)/n0) with p; Supplement A provides the exact CDF-based prediction g(p) for finite n0.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Principled model selection for stochastic dynamics." pith.science (2026). https://pith.science/paper/5SB6GJMH

@misc{pith2026250110339,
  author       = {Pith},
  title        = {Pith review of: Principled model selection for stochastic dynamics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5SB6GJMH}},
  note         = {Machine review of arXiv:2501.10339}
}
read the original abstract

Complex dynamical systems, from macromolecules to ecosystems, are often modeled by stochastic differential equations. To learn such models from data, a common approach involves sparse selection among a large function library. However, we show that overfitting arises not just from individual model complexity, but also from the combinatorial growth of possible models. To address this, we introduce Parsimonious Stochastic Inference (PASTIS), a principled method combining likelihood-estimation statistics with extreme value theory to suppress superfluous parameters. PASTIS outperforms existing methods and reliably identifies minimal models, even with low sampling rates or measurement error. It extends to stochastic partial differential equations, and applies to ecological networks and reaction-diffusion dynamics.

Figures

Figures reproduced from arXiv: 2501.10339 by the authors.

Figure 1
Figure 1. FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 4
Figure 4. Figure 4: Both rely on a Stratonovich transformation of the [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

55 extracted references · 51 canonical work pages

  1. [1]

    Inferring coefficient values

    selecting the basis functions bi(x), which is the main focus of this article, and 2) inferring the corresponding coefficient values, for which we follow an approach closely related to Stochastic Force Inference [13] (SFI). Inferring coefficient values. We first briefly sum- marize the SFI method. Our starting point is the follow- ing approximate log-likel...

    arXiv 2025

  2. [2]

    J. S. North, C. K. Wikle, and E. M. Schliep, A Review of Data-Driven Discovery for Dynamic Systems, Interna- tional Statistical Review 91, 464 (2023)

    work page 2023

  3. [3]

    B¨ ar, R

    M. B¨ ar, R. Hegger, and H. Kantz, Fitting partial differen- tial equations to space-time dynamics, Physical Review E 59, 337 (1999)

    work page 1999

  4. [4]

    J. O. Ramsay, G. Hooker, D. Campbell, and J. Cao, Pa- rameter Estimation for Differential Equations: A Gener- alized Smoothing Approach, Journal of the Royal Statis- tical Society. Series B (Statistical Methodology) 69, 741 (2007), 4623296

    work page 2007

  5. [5]

    Schmidt and H

    M. Schmidt and H. Lipson, Distilling Free-Form Natural Laws from Experimental Data, Science 324, 81 (2009)

    work page 2009

  6. [6]

    Cranmer, A

    M. Cranmer, A. Sanchez-Gonzalez, P. Battaglia, R. Xu, K. Cranmer, D. Spergel, and S. Ho, Discovering Symbolic Models from Deep Learning with Inductive Biases (2020), arXiv:2006.11287 [astro-ph, physics:physics, stat]

    arXiv 2020

  7. [7]

    S. L. Brunton, J. L. Proctor, and J. N. Kutz, Discover- ing governing equations from data by sparse identifica- tion of nonlinear dynamical systems, Proceedings of the National Academy of Sciences 113, 3932 (2016)

    2016

  8. [8]

    Champion, P

    K. Champion, P. Zheng, A. Y. Aravkin, S. L. Brunton, and J. N. Kutz, A Unified Sparse Optimization Frame- work to Learn Parsimonious Physics-Informed Models From Data, IEEE Access 8, 169259 (2020)

    work page 2020

Show all 55 references
  1. [9]

    Boninsegna, F

    L. Boninsegna, F. N¨ uske, and C. Clementi, Sparse learn- 6 ing of stochastic dynamical equations, The Journal of Chemical Physics 148, 241723 (2018)

    2018

  2. [10]

    Gao and G

    T.-T. Gao and G. Yan, Autonomous inference of com- plex network dynamics from incomplete and noisy data, Nature Computational Science 2, 160 (2022)

    2022

  3. [11]

    J. L. Callaham, J.-C. Loiseau, G. Rigas, and S. L. Brun- ton, Nonlinear stochastic modelling with Langevin re- gression, Proceedings of the Royal Society A: Mathemat- ical, Physical and Engineering Sciences 477, 20210092 (2021)

    2021

  4. [12]

    Huang, Y

    Y. Huang, Y. Mabrouk, G. Gompper, and B. Sabass, Sparse inference and active learning of stochastic differ- ential equations from data, Scientific Reports 12, 21691 (2022)

    2022

  5. [13]

    Nabeel, A

    A. Nabeel, A. Karichannavar, S. Palathingal, J. Jhawar, D. B. Br¨ uckner, D. Raj M, and V. Guttal, Discover- ing stochastic dynamical equations from ecological time series data, The American Naturalist 10.1086/734083 (2024)

    2024 doi

  6. [14]

    Frishman and P

    A. Frishman and P. Ronceray, Learning Force Fields from Stochastic Trajectories, Physical Review X 10, 021009 (2020)

    2020

  7. [15]

    H. Risken, Fokker-Planck Equation for Several Variables; Methods of Solution, in The Fokker-Planck Equation: Methods of Solution and Applications, Springer Series in Synergetics, edited by H. Risken (Springer, Berlin, Hei- delberg, 1996) pp. 133–162

    1996

  8. [16]

    Siegert, R

    S. Siegert, R. Friedrich, and J. Peinke, Analysis of data sets of stochastic systems, Physics Letters A 243, 275 (1998)

    1998

  9. [17]

    Risken, The Fokker-Planck Equation: Methods of So- lution and Applications, edited by H

    H. Risken, The Fokker-Planck Equation: Methods of So- lution and Applications, edited by H. Haken, Springer Series in Synergetics, Vol. 18 (Springer, Berlin, Heidel- berg, 1996)

    1996

  10. [18]

    P. Batz, A. Ruttor, and M. Opper, Approximate Bayes learning of stochastic differential equations, Physical Re- view E 98, 022109 (2018)

    2018

  11. [19]

    Akaike, A new look at the statistical model identifica- tion, IEEE Transactions on Automatic Control 19, 716 (1974)

    H. Akaike, A new look at the statistical model identifica- tion, IEEE Transactions on Automatic Control 19, 716 (1974)

    1974

  12. [20]

    Lebarbier and T

    ´E. Lebarbier and T. Mary-Huard, Une introduction au crit` ere BIC : fondements th´ eoriques et interpr´ etation, Journal de la soci´ et´ e fran¸ caise de statistique147, 39 (2006)

    2006

  13. [21]

    S. S. Wilks, The Large-Sample Distribution of the Likeli- hood Ratio for Testing Composite Hypotheses, The An- nals of Mathematical Statistics 9, 60 (1938)

    1938

  14. [22]

    See Supplemental Material at [URL will be inserted by publisher] for additional details on benchmarks and com- parisons, misspecified model identification, and the im- pact of the parameter p

  15. [23]

    Schwarz, Estimating the Dimension of a Model, The Annals of Statistics 6, 461 (1978), 2958889

    G. Schwarz, Estimating the Dimension of a Model, The Annals of Statistics 6, 461 (1978), 2958889

    1978

  16. [24]

    K. Aho, D. Derryberry, and T. Peterson, Model selection for ecologists: The worldviews of AIC and BIC, Ecology 95, 631 (2014), 43495189

    2014

  17. [25]

    M. T. Brolly, J. R. Maddison, A. L. Teckentrup, and J. Vanneste, Bayesian comparison of stochastic models of dispersion, Journal of Fluid Mechanics 944, A2 (2022)

    2022

  18. [26]

    A. A. Kaptanoglu, B. M. de Silva, U. Fasel, K. Kahe- man, A. J. Goldschmidt, J. Callaham, C. B. Delahunt, Z. G. Nicolaou, K. Champion, J.-C. Loiseau, J. N. Kutz, and S. L. Brunton, PySINDy: A comprehensive Python package for robust sparse system identification, Journal of Open...

    2022

  19. [27]

    Amiri, Y

    S. Amiri, Y. Zhang, A. Gerardos, C. Sykes, and P. Ronceray, Inferring geometrical dynamics of cell nu- cleus translocation, Physical Review Research 6, 043030 (2024)

    2024

  20. [28]

    Wanner and I

    M. Wanner and I. Mezi´ c, On Higher Order Drift and Diffusion Estimates for Stochastic SINDy, SIAM Journal on Applied Dynamical Systems 23, 1504 (2024)

    2024

  21. [29]

    C. L. Vestergaard, P. C. Blainey, and H. Flyvbjerg, Optimal estimation of diffusion coefficients from single- particle trajectories, Physical Review E 89, 022726 (2014)

    2014

  22. [30]

    Maddu, B

    S. Maddu, B. L. Cheeseman, I. F. Sbalzarini, and C. L. M¨ uller, Stability selection enables robust learning of dif- ferential equations from limited noisy data, Proceedings of the Royal Society A: Mathematical, Physical and En- gineering Sciences 478, 20210916 (2022)

    2022

  23. [31]

    Tibshirani, Regression Shrinkage and Selection Via the Lasso, Journal of the Royal Statistical Society Series B: Statistical Methodology 58, 267 (1996)

    R. Tibshirani, Regression Shrinkage and Selection Via the Lasso, Journal of the Royal Statistical Society Series B: Statistical Methodology 58, 267 (1996)

    1996

  24. [32]

    Ben Mhenni, S

    R. Ben Mhenni, S. Bourguignon, and J. Ninin, Global op- timization for sparse solution of least squares problems, Optimization Methods and Software 37, 1740 (2022)

    2022

  25. [33]

    N. M. Mangan, J. N. Kutz, S. L. Brunton, and J. L. Proc- tor, Model selection for dynamical systems via sparse regression and information criteria, Proceedings of the Royal Society A: Mathematical, Physical and Engineer- ing Sciences 473, 20170009 (2017)

    2017

  26. [34]

    D. B. Br¨ uckner, P. Ronceray, and C. P. Broedersz, Infer- ring the Dynamics of Underdamped Stochastic Systems, Physical Review Letters 125, 058103 (2020)

    2020

  27. [35]

    Chen and Z

    J. Chen and Z. Chen, Extended bayesian information criteria for model selection with large model spaces, Biometrika 95, 759 (2008)

    2008

  28. [36]

    J. G. Scott and J. O. Berger, Bayes and empirical-bayes multiplicity adjustment in the variable-selection problem, Annals of Statistics 38, 2587 (2010)

    2010

  29. [37]

    Y. C. Mathpati, T. Tripura, R. Nayek, and S. Chakraborty, Discovering stochastic partial dif- ferential equations from limited data using variational Bayes inference, Computer Methods in Applied Mechan- ics and Engineering 418, 116512 (2024)

    2024

  30. [38]

    The code used to produce the data and figures of this article is available at github.com/odannis/PASTIS_paper

    A Python implementation of PASTIS is available within the StochasticForceInference package at github.com/ ronceray/StochasticForceInference. The code used to produce the data and figures of this article is available at github.com/odannis/PASTIS_paper

  31. [39]

    D. B. Br¨ uckner, N. Arlt, A. Fink, P. Ronceray, J. O. R¨ adler, and C. P. Broedersz, Learning the dynamics of cell–cell interactions in confined cell migration, Pro- ceedings of the National Academy of Sciences 118, e2016602118 (2021)

    2021

  32. [40]

    Ferretti, V

    F. Ferretti, V. Chard` es, T. Mora, A. M. Walczak, and I. Giardina, Building General Langevin Models from Dis- crete Datasets, Physical Review X 10, 031018 (2020)

    2020

  33. [41]

    Rish and G

    I. Rish and G. Grabarnik, Sparse Modeling: Theory, Al- gorithms, and Applications (CRC Press, Boca Raton, 2014)

    2014

  34. [42]

    M. R. Leadbetter, G. Lindgren, and H. Rootz´ en, Asymp- totic Distributions of Extremes, in Extremes and Related Properties of Random Sequences and Processes, edited by M. R. Leadbetter, G. Lindgren, and H. Rootz´ en (Springer, New York, NY, 1983) pp. 3–30. 7

    1983

  35. [43]

    Gross and O

    E. Gross and O. Vitells, Trial factors for the look else- where effect in high energy physics, European Physical Journal C 70, 525 (2010)

    2010

  36. [44]

    Vitells and E

    O. Vitells and E. Gross, Estimating the significance of a signal in a multi-dimensional search, Astroparticle Physics 35, 230 (2011). END MA TTER

    2011

  37. [45]

    6 connecting the inference error E(ˆFB) to the estimated log-likelihood L(X|ˆFB)

    Estimating the error E from the log-likelihood We prove here Eq. 6 connecting the inference error E(ˆFB) to the estimated log-likelihood L(X|ˆFB). For sim- plicity, we assume here that the normalization matrix ¯D is equal to the exact diffusion matrix D, which has only a minor...

  38. [46]

    look-elsewhere effect

    Statistics of the information gap ∆I ∗ Here, we study the distribution of ∆ I ∗ and prove its asymptotic Gumbel distribution (Eq. 8). We use the following result from (Ref. [41], example 1.7.4): for N independent, identically distributed Gaussian random variables X1 . . . XN ∼...

  39. [47]

    9), is the probability of selecting a model with one superflu- ous term which can be written as P[maxs IPASTIS(B∗ + {s}) > IPASTIS(B∗)] ≈ p

    The parameter p in IP ASTIS Here, we show that p, present in IPASTIS(B) (Eq. 9), is the probability of selecting a model with one superflu- ous term which can be written as P[maxs IPASTIS(B∗ + {s}) > IPASTIS(B∗)] ≈ p. First, we recall that maxs IPASTIS(B∗ + {s}) − IPASTIS(B∗) ...

  40. [48]

    We derive here modified estimators discussed in the main text and presented in Fig

    Addressing Data Imperfections Large sampling intervals ∆ t and high measurement noise are two major challenges for both coefficient in- ference and model selection. We derive here modified estimators discussed in the main text and presented in Fig. 4. Both rely on a Stratonovi...

  41. [49]

    Sparse inference of stochastic partial differential equations We discuss here the adaptation of our information cri- terion IPASTIS to Stochastic Partial Differential Equa- tions (SPDEs) that we used in Fig. 3d. We consider a two-dimensional field ϕ(x, y, t) that follows: ∂ϕ(x...

  42. [50]

    For the prediction error in Fig

    Simulations details and parameters For all SDE and SPDE simulations, we use the Eu- ler–Maruyama method with simulation time interval d t and sampling time interval ∆t, and total simulation time τ . For the prediction error in Fig. 3, we simulate a new 9 independent trajectory...

  43. [51]

    3, allows comparison of models in a Bayesian framework without having to select any particular prior Π B(F B 1 ,

    Bayesian Information Criterion The Bayesian Information Criterion (BIC), against which our method is benchmarked in Fig. 3, allows comparison of models in a Bayesian framework without having to select any particular prior Π B(F B 1 , . . . , FB nB ) on the parameters F B i . I...

  44. [52]

    We note that it is primarily designed for ordinary differential equation inference, in the absence of dynamical noise

    SINDy (STLSQ) The SINDy algorithm used was based on the Python package pysindy, specifically commit master@2024-01-30 18:30:00 [S25]. We note that it is primarily designed for ordinary differential equation inference, in the absence of dynamical noise. We utilized the Sequenti...

    2024

  45. [53]

    The trajectory was divided into 7 segments of equal length

    k-F old Cross-V alidation (CV) For CV, we employed a 7-fold cross-validation scheme. The trajectory was divided into 7 segments of equal length. For each fold, 6 segments were used as the training set to learn the model parameters (i.e., the coefficients F B i for a given basi...

  46. [54]

    The LASSO regression itself was performed using the Lasso optimizer from the sklearn.linear model module in Scikit-learn

    Least Absolute Shrinkage and Selection Operator (LASSO) For LASSO, we utilized the pysindy package (commit master@2024-01-30 18:30:00) as a framework for defining the feature library and problem structure. The LASSO regression itself was performed using the Lasso optimizer fro...

    2024

  47. [55]

    S2, the comparisons in the main text (Fig

    Conclusion Based on the comprehensive hyperparameter analyses presented in Fig. S2, the comparisons in the main text (Fig. 3) are well-justified, using representative or near-optimal parameters for the benchmark methods. The figure clearly demonstrates the superior robustness ...

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.