pith. sign in

arxiv: 2605.01089 · v1 · submitted 2026-05-01 · 💻 cs.LG · math.PR· stat.CO

Learning Discriminators for Resampling in the Ensemble Gaussian Mixture Filter through a Normalizing Flow Approach

Zain Jabbar , Andrey A. Popov This is my paper

Pith reviewed 2026-05-09 19:05 UTC · model grok-4.3

classification 💻 cs.LG math.PRstat.CO
keywords ensemble Gaussian mixture filternormalizing flowsresamplingparticle filtersdata assimilationLorenz 63Ikeda mapdiscriminator learning
0
0 comments X

The pith

A normalizing flow discriminator filters out unrealistic particles during resampling to lower error in the ensemble Gaussian mixture filter when ensembles are small.

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

The paper targets a flaw in the ensemble Gaussian mixture filter where its resampling step can create particles that do not match the underlying physics, leading to poor forecasts especially with limited ensemble sizes. It adds a learned discriminator that checks each candidate particle for plausibility before acceptance. The discriminator itself is trained using a normalizing flow model on data from the target system. Tests on the Ikeda map and Lorenz 63 system show this change produces lower filtering errors than the unmodified filter in low-ensemble settings. The approach matters for practical data assimilation where computational limits keep ensemble sizes modest and unrealistic samples degrade performance.

Core claim

The discriminator-informed resampling procedure augments the posterior resampling step of the ensemble Gaussian mixture filter with a discriminator that accepts or rejects candidate particles based on physical plausibility. These discriminators are trained through a normalizing flow approach. On the Ikeda map and Lorenz 63 system the modified procedure produces consistently lower error than the standard ensemble Gaussian mixture filter in low-ensemble regimes.

What carries the argument

The discriminator-informed resampling step, in which a normalizing flow model serves as a learned gate that accepts only physically plausible particles.

If this is right

  • Unrealistic posterior samples are rejected before they enter the forecast step.
  • Filtering error decreases relative to the plain EnGMF when the ensemble size is small.
  • The improvement appears on both the Ikeda map and the Lorenz 63 system.
  • The base filter convergence properties remain unchanged while the resampling quality increases.

Where Pith is reading between the lines

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

  • The same learned discriminator could be reused across multiple forecast cycles if the underlying dynamics stay stationary.
  • Extending the discriminator to other particle filters facing similar realism problems would require only retraining the flow on the new system.
  • If the flow overfits to training trajectories, long-term forecast skill may degrade even when short-term error drops.

Load-bearing premise

A normalizing flow trained on the target system can distinguish plausible particles from implausible ones without introducing systematic bias into the filter posterior.

What would settle it

Running the discriminator-informed procedure on the Lorenz 63 system with small ensembles and finding that the error does not decrease relative to the standard EnGMF would falsify the central claim.

Figures

Figures reproduced from arXiv: 2605.01089 by Andrey A. Popov, Zain Jabbar.

Figure 1
Figure 1. Figure 1: Left: an outline of the approximated discriminator with 𝑚 = 6 along with a 3-𝜎 plot of a Gaussian distribution. Right: A 3-𝜎 plot of the discriminator-informed Gaussian distribution showing physical regions and their relative density. which defines the inverse map as,  −1(𝑥1,𝑛+1, 𝑥2,𝑛+1) = (𝑥1,𝑛, 𝑥,𝑛), (17) and constitutes an exact inverse of the forward map eq. (13). Using this inverse, a family of discr… view at source ↗
Figure 2
Figure 2. Figure 2: For the Ikeda map. For a bandwidth scaling factor eq. (9) of 𝑠𝛽 = 1: A comparison of the RMSE eq. (37) for the DI-EnGMF (square markers and blue dotted line), EnGMF (triangle markers and orange dash-dotted line), and the EnKF (circle markers with green solid line) for ensemble sizes in the range 𝑁 = 3 to 𝑁 = 20. The value plotted is the mean over all Monte Carlo iterations and the vertical lines represent … view at source ↗
Figure 3
Figure 3. Figure 3: For the Lorenz ’63 equations, a comparison of the RMSE eq. (37) for the DI-EnGMF (triangle markers and green solid line), EnGMF (circle markers and blue dotted line), and the EnKF (square markers with orange dash-dotted line) for ensemble sizes in the range 𝑁 = 10 to 𝑁 = 200. The value plotted is the mean over all Monte Carlo iterations and the vertical lines represent three standard errors of the mean (SE… view at source ↗
read the original abstract

The ensemble Gaussian mixture filter (EnGMF) is a powerful, convergent particle filter capable of medium-to-high dimensional non-linear filtering. The EnGMF relies on a resampling step that can generate physically unrealistic posterior samples, that would subsequently produce physically meaningless forecasts. This work introduces the discriminator-informed resampling procedure, that augments the posterior resampling step with a discriminator that accepts or rejects candidate particles based on their physical plausibility. In this work these discriminators are learned through a normalizing flow approach. Numerical experiments on both the Ikeda map and the Lorenz '63 system show that discriminator informed resampling procedure consistently reduces error relative to the standard EnGMF in low-ensemble regimes.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript introduces a discriminator-informed resampling procedure for the Ensemble Gaussian Mixture Filter (EnGMF), where a discriminator learned via normalizing flows accepts or rejects candidate particles based on physical plausibility. This is intended to prevent the generation of physically unrealistic posterior samples during resampling. Numerical experiments on the Ikeda map and Lorenz '63 system demonstrate that this procedure consistently reduces error relative to the standard EnGMF in low-ensemble regimes.

Significance. If the method maintains the unbiasedness and convergence properties of the EnGMF while effectively filtering out implausible particles, it represents a promising hybrid approach combining data-driven discriminators with traditional particle filters. This could have significant implications for improving the performance of ensemble-based data assimilation in nonlinear dynamical systems, particularly when ensemble sizes are limited due to computational constraints.

major comments (2)
  1. [Numerical experiments] The claim of consistent error reduction is not supported by any quantitative metrics, specific ensemble sizes, training details for the normalizing flow, or statistical tests. Without these, it is impossible to evaluate the practical significance or reproducibility of the reported improvements on the Ikeda map and Lorenz '63 system.
  2. [Theoretical analysis] The manuscript does not provide a proof or argument that the acceptance/rejection step using the learned discriminator preserves the posterior distribution or the convergence guarantees of the original EnGMF. This is load-bearing because any systematic bias introduced by the discriminator could invalidate the filter's theoretical properties and explain the observed error reduction as an artifact rather than a true improvement.
minor comments (1)
  1. [Abstract] The abstract could benefit from a brief mention of the specific normalizing flow architecture used or the training procedure to give readers a better sense of the method's implementation.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for their detailed review and constructive suggestions. We address each major comment below and outline the revisions we will make to the manuscript.

read point-by-point responses

  1. Referee: [Numerical experiments] The claim of consistent error reduction is not supported by any quantitative metrics, specific ensemble sizes, training details for the normalizing flow, or statistical tests. Without these, it is impossible to evaluate the practical significance or reproducibility of the reported improvements on the Ikeda map and Lorenz '63 system.

    Authors: We agree that additional quantitative details would strengthen the presentation. In the revised version, we will add a table summarizing the root mean square error (RMSE) reductions for specific ensemble sizes (N=10, 20, 50) on both the Ikeda map and Lorenz '63 systems. We will also include details on the normalizing flow architecture (e.g., number of layers, hidden units), training procedure (epochs, batch size, optimizer), and results from multiple independent runs with statistical significance tests (e.g., Wilcoxon signed-rank test) to confirm the improvements are consistent and not due to random variation. The current manuscript relies on visual comparisons in the figures, but we will enhance this with the requested metrics. revision: yes

  2. Referee: [Theoretical analysis] The manuscript does not provide a proof or argument that the acceptance/rejection step using the learned discriminator preserves the posterior distribution or the convergence guarantees of the original EnGMF. This is load-bearing because any systematic bias introduced by the discriminator could invalidate the filter's theoretical properties and explain the observed error reduction as an artifact rather than a true improvement.

    Authors: This is a valid concern. The EnGMF resampling generates particles from a Gaussian mixture model approximating the posterior, which can occasionally produce unphysical samples. Our discriminator, trained via normalizing flows to model the distribution of physically plausible states, is intended to reject such outliers. While we do not provide a formal proof that this exactly preserves the posterior (as the discriminator is an approximation), we will add a new subsection in the revised manuscript providing an argument based on the properties of acceptance-rejection sampling: if the discriminator accurately approximates the indicator function for the support of the true posterior, the procedure remains unbiased in the limit of perfect discrimination. We will also discuss the potential for bias in finite-sample cases and note that the empirical improvements suggest the bias, if present, is outweighed by the variance reduction. A full theoretical analysis of convergence is beyond the scope of this work but is planned for future research. revision: partial

standing simulated objections not resolved
  • Rigorous proof that the acceptance/rejection step preserves the posterior distribution and convergence guarantees of the EnGMF

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper augments the EnGMF with a learned discriminator via normalizing flows for resampling and validates the approach through numerical experiments on the Ikeda map and Lorenz '63 system. No derivation chain, equation, or prediction reduces by construction to a fitted input or self-citation; the error-reduction claim rests on external empirical benchmarks rather than tautological redefinition of the method's own outputs. The base EnGMF convergence properties are treated as given from prior literature without the new component being forced by self-reference.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The abstract does not disclose any free parameters, axioms, or invented entities beyond the standard components of the EnGMF and normalizing flows; the discriminator is learned rather than postulated as a new physical entity.

pith-pipeline@v0.9.0 · 5414 in / 1071 out tokens · 32967 ms · 2026-05-09T19:05:00.973565+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

108 extracted references · 108 canonical work pages

  1. [1]

    1986 , publisher=

    Density estimation for statistics and data analysis , author=. 1986 , publisher=

    work page 1986

  2. [2]

    Biostatistics & Epidemiology , volume=

    A tutorial on kernel density estimation and recent advances , author=. Biostatistics & Epidemiology , volume=. 2017 , publisher=

    work page 2017

  3. [3]

    Janssen, J

    P. Janssen and J. S. Marron and N. Veraverbeke and W. Sarle , doi =. Scale measures for bandwidth selection , url =. 1995 , bdsk-url-1 =. https://doi.org/10.1080/10485259508832654 , journal =

    work page doi:10.1080/10485259508832654 1995

  4. [4]

    Technometrics , volume=

    Variable kernel estimates of multivariate densities , author=. Technometrics , volume=. 1977 , publisher=

    work page 1977

  5. [5]

    IEEE Transactions on Aerospace and Electronic Systems , volume=

    Clustering Methods for Particle Filters with Gaussian Mixture Models , author=. IEEE Transactions on Aerospace and Electronic Systems , volume=. 2021 , publisher=

    work page 2021

  6. [6]

    IEEE Transactions on Signal Processing , volume=

    Gaussian mixture nonlinear filtering with resampling for mixand narrowing , author=. IEEE Transactions on Signal Processing , volume=. 2016 , publisher=

    work page 2016

  7. [7]

    Journal of Guidance, Control, and Dynamics , volume=

    Gaussian sum reapproximation for use in a nonlinear filter , author=. Journal of Guidance, Control, and Dynamics , volume=. 2015 , publisher=

    work page 2015

  8. [8]

    SIAM Journal on Scientific Computing , volume=

    A multifidelity ensemble Kalman filter with reduced order control variates , author=. SIAM Journal on Scientific Computing , volume=. 2021 , publisher=

    work page 2021

  9. [9]

    Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications (Vol

    Multifidelity Data Assimilation for Physical Systems , author=. Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications (Vol. IV) , pages=. 2022 , publisher=

    work page 2022

  10. [10]

    Bridging the ensemble Kalman filter and particle filters: the adaptive

    Stordal, Andreas S and Karlsen, Hans A and N. Bridging the ensemble Kalman filter and particle filters: the adaptive. Computational Geosciences , volume=. 2011 , publisher=

    work page 2011

  11. [11]

    Quarterly Journal of the Royal Meteorological Society , volume=

    Particle filters for high-dimensional geoscience applications: A review , author=. Quarterly Journal of the Royal Meteorological Society , volume=. 2019 , publisher=

    work page 2019

  12. [12]

    Anderson, Jeffrey L and Anderson, Stephen L , journal=. A

    work page

  13. [13]

    Efficient kernel-based ensemble

    Liu, Bo and Ait-El-Fquih, Boujemaa and Hoteit, Ibrahim , journal=. Efficient kernel-based ensemble

    work page

  14. [14]

    Kernel-based ensemble

    Yun, Sehyun and Zanetti, Renato and Jones, Brandon A , journal=. Kernel-based ensemble. 2022 , publisher=

    work page 2022

  15. [15]

    An adaptive covariance parameterization technique for the ensemble

    Popov, Andrey A and Zanetti, Renato , journal=. An adaptive covariance parameterization technique for the ensemble. 2024 , publisher=

    work page 2024

  16. [16]

    The Ensemble

    Popov, Andrey A and Zanetti, Renato , journal=. The Ensemble

    work page

  17. [17]

    and Popov, Andrey A

    Reifler, Benjamin L. and Popov, Andrey A. and Jones, Brandon A. and Zanetti, Renato , journal =. Large-scale space object tracking in a proliferated

    work page

  18. [18]

    and Zanetti, Renato , booktitle=

    Popov, Andrey A. and Zanetti, Renato , booktitle=. Ensemble. 2023 , pages=

    work page 2023

  19. [19]

    Ensemble-localized Kernel Density Estimation with Applications to the Ensemble

    Popov, Andrey A and Zanetti, Renato , journal=. Ensemble-localized Kernel Density Estimation with Applications to the Ensemble

    work page

  20. [20]

    Multi-target ensemble

    Reifler, Benjamin L and Yun, Sehyun and Jones, Brandon A and Zanetti, Renato , booktitle=. Multi-target ensemble

    work page

  21. [21]

    and Popov, Andrey A

    Reifler, Benjamin L. and Popov, Andrey A. and Jones, Brandon A. and Zanetti, Renato , booktitle =. Large-scale space object tracking in a proliferated

    work page

  22. [22]

    Nonlinear Bayesian estimation using

    Alspach, Daniel and Sorenson, Harold , journal=. Nonlinear Bayesian estimation using. 1972 , publisher=

    work page 1972

  23. [23]

    Automatica , volume=

    Recursive Bayesian estimation using Gaussian sums , author=. Automatica , volume=. 1971 , publisher=

    work page 1971

  24. [24]

    Journal of guidance, control, and dynamics , volume=

    Uncertainty propagation for nonlinear dynamic systems using Gaussian mixture models , author=. Journal of guidance, control, and dynamics , volume=

    work page

  25. [25]

    IEEE Signal Processing Letters , volume=

    The split and merge unscented Gaussian mixture filter , author=. IEEE Signal Processing Letters , volume=. 2009 , publisher=

    work page 2009

  26. [26]

    What are You Weighting For? Improved Weights for

    Durant, Dalton and Popov, Andrey A and Zanetti, Renato , journal=. What are You Weighting For? Improved Weights for

    work page

  27. [27]

    and Zanetti, Renato , booktitle=

    Popov, Andrey A. and Zanetti, Renato , booktitle=. Are Non-Gaussian Kernels Suitable for Ensemble Mixture Model Filtering? , day =

    work page

  28. [28]

    Durant, Dalton and Popov, Andrey A and Zanetti, Renato , booktitle=

    work page

  29. [29]

    IEEE Transactions on Aerospace and Electronic Systems , year=

    Gaussian Mixture-Based Point Mass Filtering With Applications to Terrain-Relative Navigation , author=. IEEE Transactions on Aerospace and Electronic Systems , year=

    work page

  30. [30]

    Ensemble Gaussian Mixture Filter based on Projected Cram

    Hanebeck, Uwe D and Prossel, Dominik and Popov, Andrey A and Giraldo-Grueso, Felipe and Zanetti, Renato , booktitle=. Ensemble Gaussian Mixture Filter based on Projected Cram. 2025 , organization=

    work page 2025

  31. [31]

    2019 , publisher=

    Wang, Ziqi and Broccardo, Marco and Song, Junho , journal=. 2019 , publisher=

    work page 2019

  32. [32]

    2013 , publisher=

    A catalog of special plane curves , author=. 2013 , publisher=

    work page 2013

  33. [33]

    Bayesian inference of chaotic dynamics by merging data assimilation, machine learning and expectation- maximization.arXiv preprint arXiv:2001.06270,

    Bayesian inference of chaotic dynamics by merging data assimilation, machine learning and expectation-maximization , author=. arXiv preprint arXiv:2001.06270 , year=

    work page arXiv 2001

  34. [34]

    A view of the

    Neal, Radford M and Hinton, Geoffrey E , booktitle=. A view of the. 1998 , publisher=

    work page 1998

  35. [35]

    2006 , publisher=

    Pattern recognition and machine learning , author=. 2006 , publisher=

    work page 2006

  36. [36]

    2012 , publisher=

    Optimal filtering , author=. 2012 , publisher=

    work page 2012

  37. [37]

    2004 , publisher=

    Estimation with applications to tracking and navigation: theory algorithms and software , author=. 2004 , publisher=

    work page 2004

  38. [38]

    Gromov's method for

    Daum, Fred and Huang, Jim and Noushin, Arjang , booktitle=. Gromov's method for. 2016 , organization=

    work page 2016

  39. [39]

    IEEE Transactions on signal processing , volume=

    Particle filters for positioning, navigation, and tracking , author=. IEEE Transactions on signal processing , volume=. 2002 , publisher=

    work page 2002

  40. [40]

    2016 , publisher=

    Data assimilation: methods, algorithms, and applications , author=. 2016 , publisher=

    work page 2016

  41. [41]

    2015 , publisher=

    Probabilistic forecasting and Bayesian data assimilation , author=. 2015 , publisher=

    work page 2015

  42. [42]

    Nonlinear Processes in Geophysics , volume=

    A Bayesian approach to multivariate adaptive localization in ensemble-based data assimilation with time-dependent extensions , author=. Nonlinear Processes in Geophysics , volume=. 2019 , publisher=

    work page 2019

  43. [43]

    Biometrika , volume=

    Bridging the ensemble Kalman and particle filters , author=. Biometrika , volume=. 2013 , publisher=

    work page 2013

  44. [44]

    A nonparametric ensemble transform method for

    Reich, Sebastian , journal=. A nonparametric ensemble transform method for. 2013 , publisher=

    work page 2013

  45. [45]

    arXiv preprint arXiv:2003.13162 , year=

    An explicit probabilistic derivation of inflation in a scalar ensemble Kalman filter for finite step, finite ensemble convergence , author=. arXiv preprint arXiv:2003.13162 , year=

    work page arXiv 2003

  46. [46]

    Ensemble

    Michaelson, Kristen and Popov, Andrey A and Zanetti, Renato , booktitle=. Ensemble. 2023 , organization=

    work page 2023

  47. [47]

    Proceedings of the 33rd AAS/AIAA Space Flight Mechanics Meeting , year=

    Recursive update filtering: A new approach , author=. Proceedings of the 33rd AAS/AIAA Space Flight Mechanics Meeting , year=

    work page

  48. [48]

    2024 27th International Conference on Information Fusion (FUSION) , pages=

    Particle flow with a continuous formulation of the nonlinear measurement update , author=. 2024 27th International Conference on Information Fusion (FUSION) , pages=. 2024 , organization=

    work page 2024

  49. [49]

    2007 , publisher=

    Stochastic processes and filtering theory , author=. 2007 , publisher=

    work page 2007

  50. [50]

    Ocean Dynamics , volume=

    Ensemble Kalman filter implementations based on shrinkage covariance matrix estimation , author=. Ocean Dynamics , volume=. 2015 , publisher=

    work page 2015

  51. [51]

    Nonlinear Processes in Geophysics , volume=

    A stochastic covariance shrinkage approach to particle rejuvenation in the ensemble transform particle filter , author=. Nonlinear Processes in Geophysics , volume=. 2022 , publisher=

    work page 2022

  52. [52]

    A stochastic covariance shrinkage approach in ensemble transform

    Popov, Andrey A and Sandu, Adrian and Nino-Ruiz, Elias D and Evensen, Geir , journal=. A stochastic covariance shrinkage approach in ensemble transform

    work page

  53. [53]

    2009 ieee international conference on acoustics, speech and signal processing , pages=

    Shrinkage estimation of high dimensional covariance matrices , author=. 2009 ieee international conference on acoustics, speech and signal processing , pages=. 2009 , organization=

    work page 2009

  54. [54]

    Shrinkage algorithms for MMSE covariance estimation , Volume =

    Chen, Yilun and Wiesel, Ami and Eldar, Yonina C and Hero, Alfred O , Journal =. Shrinkage algorithms for MMSE covariance estimation , Volume =

    work page

  55. [55]

    Shrinkage-to-tapering estimation of large covariance matrices , Volume =

    Chen, Xiaohui and Wang, Z Jane and McKeown, Martin J , Journal =. Shrinkage-to-tapering estimation of large covariance matrices , Volume =

    work page

  56. [56]

    Robust shrinkage estimation of high-dimensional covariance matrices , Volume =

    Chen, Yilun and Wiesel, Ami and Hero, Alfred O , Journal =. Robust shrinkage estimation of high-dimensional covariance matrices , Volume =

    work page

  57. [57]

    A well conditioned estimator for large dimensional covariance matrices , Url =

    Ledoit, Olivier and Wolf, Michael , Date-Modified =. A well conditioned estimator for large dimensional covariance matrices , Url =. Journal of Multivariate Analysis , Number =

    work page

  58. [58]

    Society for industrial and applied mathematics, undergraduate research online , pages=

    Attractors: Nonstrange to chaotic , author=. Society for industrial and applied mathematics, undergraduate research online , pages=

    work page

  59. [59]

    IEEE transactions on pattern analysis and machine intelligence , volume=

    Normalizing flows: An introduction and review of current methods , author=. IEEE transactions on pattern analysis and machine intelligence , volume=. 2020 , publisher=

    work page 2020

  60. [60]

    Advances in neural information processing systems , volume=

    Neural spline flows , author=. Advances in neural information processing systems , volume=

    work page

  61. [61]

    A new approach to linear filtering and prediction problems , author=

    work page

  62. [62]

    2007 , publisher=

    Statistical multisource-multitarget information fusion , author=. 2007 , publisher=

    work page 2007

  63. [63]

    Computers & Mathematics with Applications , volume=

    Runge--Kutta pairs of order 5 (4) satisfying only the first column simplifying assumption , author=. Computers & Mathematics with Applications , volume=. 2011 , publisher=

    work page 2011

  64. [64]

    2008 , publisher=

    Fundamentals of Stochastic Filtering , author=. 2008 , publisher=

    work page 2008

  65. [65]

    2006 , publisher=

    Dynamical systems, graphs, and algorithms , author=. 2006 , publisher=

    work page 2006

  66. [66]

    Chaos: An Interdisciplinary Journal of Nonlinear Science , volume=

    Mapping chaos: Bifurcation patterns and shrimp structures in the Ikeda map , author=. Chaos: An Interdisciplinary Journal of Nonlinear Science , volume=. 2024 , publisher=

    work page 2024

  67. [67]

    Optics communications , volume=

    Multiple-valued stationary state and its instability of the transmitted light by a ring cavity system , author=. Optics communications , volume=. 1979 , publisher=

    work page 1979

  68. [68]

    Physical Review Letters , volume=

    Optical turbulence: chaotic behavior of transmitted light from a ring cavity , author=. Physical Review Letters , volume=. 1980 , publisher=

    work page 1980

  69. [69]

    2021 , publisher=

    Ordinary differential equations and dynamical systems , author=. 2021 , publisher=

    work page 2021

  70. [70]

    SIAM Journal on control and Optimization , volume=

    Projected Newton methods for optimization problems with simple constraints , author=. SIAM Journal on control and Optimization , volume=. 1982 , publisher=

    work page 1982

  71. [71]

    2003 , publisher=

    Solving nonlinear equations with Newton's method , author=. 2003 , publisher=

    work page 2003

  72. [72]

    Uninformative priors for

    Price, Harold J and Manson, Allison R , booktitle=. Uninformative priors for. 2002 , organization=

    work page 2002

  73. [73]

    Predictability: A problem partly solved , author=. Proc. Seminar on predictability , volume=. 1996 , organization=

    work page 1996

  74. [74]

    Novel approach to nonlinear/non-

    Gordon, Neil J and Salmond, David J and Smith, Adrian FM , booktitle=. Novel approach to nonlinear/non-. 1993 , organization=

    work page 1993

  75. [75]

    Advances in neural information processing systems , volume=

    Generative adversarial nets , author=. Advances in neural information processing systems , volume=

    work page

  76. [76]

    2013 , publisher=

    Auto-encoding variational bayes , author=. 2013 , publisher=

    work page 2013

  77. [77]

    Journal of Machine Learning Research , volume=

    Normalizing flows for probabilistic modeling and inference , author=. Journal of Machine Learning Research , volume=

    work page

  78. [78]

    Physica D: nonlinear phenomena , volume=

    Measuring the strangeness of strange attractors , author=. Physica D: nonlinear phenomena , volume=. 1983 , publisher=

    work page 1983

  79. [79]

    Lecture notes in mathematics , volume=

    Functional differential equations and approximation of fixed points , author=. Lecture notes in mathematics , volume=. 1979 , publisher=

    work page 1979

  80. [80]

    Mathematica in Action: Problem Solving Through Visualization and Computation , pages=

    Computational Geometry , author=. Mathematica in Action: Problem Solving Through Visualization and Computation , pages=. 2010 , publisher=

    work page 2010

Showing first 80 references.