Pith. sign in

REVIEW 4 major objections 4 minor 79 references

Full Domain Analysis in Fluid Dynamics

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper introduces full domain analysis, a framework for mapping the entire space of high-performing designs in expensive fluid-dynamics problems, and demonstrates it by generating a million flow shapes from a thousand simulations.

desk verdict A tidy framework synthesis with a well-integrated demo, but the headline 'insights' are unvalidated surrogate predictions and the one direct check already contradicts a prediction. read the letter →

arxiv 2505.22275 v1 pith:TQEQAS5F submitted 2025-05-28 cs.LG cs.NE

Alexander Hagg , Adam Gaier , Dominik Wilde , Alexander Asteroth , Holger Foysi , Dirk Reith This is my paper
classification cs.LGcs.NE
keywords fulldomainanalysisqualitydiversitysurrogatemodelsvariationalautoencoderlatticeBoltzmannshapeencodingdesignspaceexplorationcomputationalfluiddynamics
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper introduces full domain analysis, a framework for understanding the full space of solutions in an expensive computational domain rather than searching for a single optimum. It argues that combining shape encodings, quality-diversity search, fast CFD, surrogate models, and interactive visualization lets engineers efficiently generate, analyze, and refine diverse high-performing designs. The demonstration on 2D building footprints produces a million solutions from a thousand simulations and recovers correlations between footprint area, enstrophy, and maximum flow velocity. The core claim is that this structured, iterative loop is feasible today with existing component methods and gives engineers a practical path from design definition to domain insight.

What carries the argument

The machinery is the FDA loop itself: a parametric encoding maps a low-dimensional search space to shapes in a reachable manifold; quality-diversity search, specifically SPHEN, fills an archive whose niches are defined by morphological and flow features; Gaussian-process surrogates with a squared-exponential kernel predict fitness and features so that not every candidate needs a CFD run; a GPU-based lattice Boltzmann solver, Lettuce, evaluates the selected shapes; and a convolutional VAE compresses the resulting shape set into a five-dimensional latent space for interactive exploration. The GP surrogate is the component that makes the million-solution analysis possible, and it is also the point where predicted and simulated values diverge.

What would settle it

Take a random sample of, say, 100 shapes from the 1,000,000 generated solutions, run full lattice Boltzmann simulations on each, and compare the simulated area, enstrophy, and $u_{\max}$ against the GP predictions; if the error distribution is large or the enstrophy-velocity correlation flips under simulation, the quantitative conclusions of the FDA demonstration fail.

Watch

Extended reading notes

Core claim

The paper's central claim is that full domain analysis is a viable alternative to single-solution optimization in fluid dynamics, and that its components—encodings, divergent search, CFD, efficiency, and visualization—can be combined into an effective iterative loop. In the demonstration, a spline encoding of eight control points is searched by surrogate-assisted phenotypic niching, which trains Gaussian-process surrogates on 1,000 Lettuce lattice Boltzmann simulations and uses them to evaluate 2,250,000 proposed solutions. A convolutional VAE then compresses the archive into a five-dimensional latent space, enabling the generation of 1,000,000 new shapes whose area, enstrophy, and maximum velocity are predicted by the same surrogates. The paper reports an almost linear enstrophy-versus-velocity relationship and shows a user selecting a shape with predicted lower maximum velocity, which is then validated by simulation; the simulated value of $u_{\max} = 2.49$ is higher than predicted, indicating the surrogate overpredicted performance.

Load-bearing premise

The whole analysis depends on the Gaussian-process surrogates staying accurate across the latent space; if they misjudge the flow around shapes they have not seen, the correlations and rankings derived from a million predicted solutions are not trustworthy.

Editorial extensions

If this is right

  • Engineers can shift from finding one optimum to exploring all plausible designs, reducing the risk of early design decisions in expensive domains like the built environment.
  • The same pipeline of encoding, divergent search, surrogate prediction, and latent visualization can transfer to other expensive simulation domains beyond fluids, such as structural or thermal design.
  • Correlations between morphology and flow features, such as enstrophy and maximum velocity, become discoverable from data rather than assumed from prior knowledge.
  • A user can zoom into design regions, retrain the generative model on selected shapes, and re-run the loop, turning FDA into an interactive design tool rather than a one-shot optimizer.

Reading between the lines

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

  • The reported correlations, including the enstrophy-versus-velocity relationship, are computed largely from surrogate outputs; because the one direct validation shows the surrogate underpredicted $u_{\max}$, the million-solution statistics should be re-checked with sampled simulations before they guide design decisions.
  • The VAE's five latent dimensions are treated as morphological features, but without a disentanglement evaluation their correspondence to interpretable shape properties remains to be demonstrated.
  • The framework should extend naturally to 3D and higher Reynolds number flows, but the stability-accuracy tradeoff in unsupervised CFD will likely force stronger reliance on surrogates and mesh-free methods.
  • A testable extension would be to feed the correlation structure discovered in the latent space back into the QD archive's feature definitions, closing the loop between data-driven and hand-defined diversity.
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 / 4 minor

Summary. The paper introduces 'Full Domain Analysis' (FDA), a methodological framework for efficiently exploring and analyzing the full space of solutions in expensive simulation domains, with fluid dynamics as the running example. It proposes five components—encodings, search, CFD, efficiency, and visualization—and reviews state-of-the-art methods for each, including direct/indirect/latent encodings, quality-diversity search, surrogate-assisted efficiency, and latent generative models. The paper then presents a demonstration: a 2D flow problem around building-footprint spline shapes, using SPHEN with Gaussian process surrogates trained on 1,000 Lettuce simulations to populate a 1,000-solution archive, a VAE trained on 4,000 solutions, and an analysis of 1,000,000 predicted solutions. The headline findings are that enstrophy and maximum flow velocity are positively correlated, that larger area leads to higher maximum velocity, and that the VAE latent space allows interactive morphological and flow-feature analysis.

Significance. If the framework and demonstration hold, FDA would be a useful synthesis for engineering design exploration: it gives a structured, iterative process for generating diverse solution sets, predicting behavior with surrogates, and interactively navigating expensive simulation domains. The paper's strengths are its comprehensive literature review, the concrete implementation of the entire pipeline (Lettuce, SPHEN, VAE, GP surrogates), and a clearly described end-to-end case study. However, the demonstration's central 'what we can learn' analyses are built entirely on surrogate predictions without error quantification, and the single direct validation contradicts a prediction, so the current evidence does not yet establish that the reported correlations are trustworthy.

major comments (4)
  1. [§6.5, §6.6, Fig. 11] The central demonstration treats surrogate predictions as ground truth. The 1,000,000-solution analysis in §6.6 and the isolines in Fig. 11 are computed from Gaussian process outputs, not from Lettuce simulations, yet no error bars, calibration plot, or validation set of solver runs is reported. The only direct validation, in §6.5, states that the actual umax = 2.49 is higher than predicted, which is a quantified discrepancy with no discussion of its implications. Without an assessment of surrogate error across the VAE latent space, the claimed near-linear enstrophy-umax relationship and the isoline structure are not established.
  2. [§6.2–§6.4] The reported correlations are partly circular. The GP models are trained on 1,000 Lettuce samples from the same domain, the SPHEN archive is populated using those GPs, the VAE is trained on 4,000 solutions drawn from that archive, and the million-solution analysis in §6.6 again uses GP outputs for VAE-decoded shapes. The observed enstrophy-umax signal is therefore inherited from the training data and the surrogate-guided selection process rather than independently discovered. The paper should either validate the headline correlations on an independent set of Lettuce simulations or explicitly reframe them as statements about the surrogate-consistent portion of the domain.
  3. [§2–§3, Eqs. (2)–(3)] The abstract promises a formal model of full domain analysis, but the optimization-theoretic core in Eqs. (2)–(3) depends on a niche threshold epsilon that the text itself says is domain-dependent and possibly non-constant; no method for setting or interpreting epsilon is provided. Likewise, the encoding requirements in §2.1 are stated as qualitative desiderata rather than measurable properties. For a framework paper this level of precision may be acceptable, but the claim of a 'formal model' overstates the current rigor and should be tempered or the model should be made operational.
  4. [§6.2] The efficiency statement 'we were able to evaluate, in a surrogate-assisted manner, 2,250,000 proposed solutions' is not supported by the described procedure. After 1,000 archive updates with 25 new solutions per update, the total number of created candidates is 25,000; the origin of the factor 90 is unexplained. This number should be corrected, or the procedure description should be expanded so that the claimed total is reproducible.
minor comments (4)
  1. [§6.1] The sentence 'see Fig. 6. The bitmaps are used...' is missing a formatting space and reads as if Fig. 6 itself is the sentence; it should be 'see Fig. 6. The bitmaps...'.
  2. [Eq. (4)] The squared exponential covariance function is written for scalar x and x' despite x being a multi-dimensional parameter vector; the equation should use a norm, e.g., ||x - x'||^2.
  3. [Fig. 7 and §6.2/§6.4] The archive size is stated as 1,000 in §6.2 and Fig. 7a but as 4,000 in Fig. 7a's caption and §6.4; the relationship between these two archives should be clarified.
  4. [§6.5] The claim that 'E and umax are positively related' is presented as a finding from visual inspection of surrogate predictions; adding a quantitative correlation measure with uncertainty would strengthen this claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the framework contribution is self-contained, and the surrogate-based demonstration is an honest, if unvalidated, illustration rather than a circular derivation.

full rationale

The paper's central claim is a framework/requirement model for full-domain analysis, not a derivation of a physical result. The demonstration re-runs previously published components (SPHEN with 1000 Lettuce samples, a VAE, and GP surrogates), and the self-citations to Hagg et al. (2020b), Gaier et al. (2018a), and Hagg et al. (2020c) are prior-art references rather than imported uniqueness theorems. The apparent concern that Fig. 11's E-umax relationship is produced by GP surrogates fitted to the same Lettuce data is a validity limitation, not a circular reduction: the paper explicitly labels the archive as "predicted" (Section 6.2), states the expected correlation ("Of course we expect higher umax to lead to higher E"), and reports the single direct simulation check with a mismatch ("The actual value of umax = 2.49 is higher than predicted"). The 1,000,000-point plot is an interpolation from the fitted model, not an independent confirmation, but it is not presented as a first-principles result or as an external benchmark. The framework's definitions and component requirements are given independently of the demo's empirical findings, and no equation in the paper reduces by construction to its own input. Therefore no significant circularity is present; the surrogate-validation weakness belongs to correctness/robustness, not to circular reasoning.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard machine learning assumptions (GP smoothness, VAE latent continuity) and on hand-chosen hyperparameters that are not systematically varied. The framework decomposition itself is an ad hoc taxonomy introduced in this paper. No physical entities are invented.

free parameters (5)
  • GP kernel length scale l and signal variance sigma = not reported
    Fit by maximizing log-likelihood with GPML (Section 6.2); they define the surrogate model that produces all predictions used in the million-solution analysis.
  • QD perturbation standard deviation = sigma = 0.1
    Hand-chosen mutation size for generating new solutions in SPHEN (Section 6.2).
  • Archive size = 1000 (and 4000)
    Hand-chosen capacity of the Voronoi archive; affects which solutions are kept (Sections 6.2, 6.4).
  • VAE latent dimensionality = 5
    Hand-chosen number of latent variables; determines the morphological search space (Section 6.4).
  • Niche threshold epsilon in Eqs. (2)-(3) = not specified
    Domain-dependent distance threshold for defining local optima or niches; never assigned a concrete value in the demonstration.
assumptions (4)
  • standard math Gaussian process regression with a squared exponential kernel models the mapping from shape parameters to flow features smoothly.
    Invoked in Section 6.2; the GP surrogate is the backbone of the efficient search and all later predictions.
  • domain assumption 2D lattice Boltzmann simulation at Re=3900 and Mach=0.075 is a valid proxy for wind nuisance around real high-rise building footprints.
    Section 6.1 states the domain is close to real world problems but is a simplified 2D model.
  • domain assumption The VAE latent space is regularized to be smooth and continuous, so interpolations and GP predictions on latent coordinates are meaningful.
    Section 6.5 relies on the latent space as a search space; no quantitative check of smoothness is provided.
  • ad hoc to paper The five components (encoding, search, CFD, efficiency, visualization) are sufficient and necessary to define full domain analysis.
    This decomposition is asserted in Section 1 without argument or comparison with other taxonomies.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Full Domain Analysis in Fluid Dynamics." pith.science (2026). https://pith.science/paper/TQEQAS5F

@misc{pith2026250522275,
  author       = {Pith},
  title        = {Pith review of: Full Domain Analysis in Fluid Dynamics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TQEQAS5F}},
  note         = {Machine review of arXiv:2505.22275}
}
read the original abstract

Novel techniques in evolutionary optimization, simulation and machine learning allow for a broad analysis of domains like fluid dynamics, in which computation is expensive and flow behavior is complex. Under the term of full domain analysis we understand the ability to efficiently determine the full space of solutions in a problem domain, and analyze the behavior of those solutions in an accessible and interactive manner. The goal of full domain analysis is to deepen our understanding of domains by generating many examples of flow, their diversification, optimization and analysis. We define a formal model for full domain analysis, its current state of the art, and requirements of subcomponents. Finally, an example is given to show what we can learn by using full domain analysis. Full domain analysis, rooted in optimization and machine learning, can be a helpful tool in understanding complex systems in computational physics and beyond.

Figures

Figures reproduced from arXiv: 2505.22275 by the authors.

Figure 1
Figure 1. User process perspective on FDA. After the user defines the domain through available initial [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. While search and optimization takes place in which we call the parameter space [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Three main categories of shape encodings to produce solutions in [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Heterogeneous fitness landscapes often contain clusters of varying sizes, making the definition of [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Multi-solution optimization. Multi-objective optimization (a) finds a Pareto front of trade-off solutions. Solutions are added to the front if they dominate neighboring solutions in at least one objective. In multimodal optimization (b), solutions are selected through …
Figure 6
Figure 6. Figure 6: Encoding of 2D shapes: eight control points’ polar coordinates. [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]
Figure 7
Figure 7. Figure 7: After efficiently training surrogate models for the features (area and enstrophy/turbulence) and [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Architecture of the convolutional VAE generative model. [PITH_FULL_IMAGE:figures/full_fig_p015_8.png]
Figure 9
Figure 9. Figure 9: The three features are shown for a shape and its variants, when varying five latent dimensions in [PITH_FULL_IMAGE:figures/full_fig_p015_9.png]
Figure 10
Figure 10. Figure 10: Flow around selected shape. The top row shows the appearance of a strong vortex (white outline) [PITH_FULL_IMAGE:figures/full_fig_p016_10.png]
Figure 11
Figure 11. Figure 11: 1,000,000 solutions generated by VAE (a), subselection of 10,000 shapes (b), isolines of min, mean [PITH_FULL_IMAGE:figures/full_fig_p016_11.png]
Figure 12
Figure 12. Figure 12: FDA implementation example. First, the QD process is bootstrapped with a predefined spline [PITH_FULL_IMAGE:figures/full_fig_p017_12.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

79 extracted references · 78 canonical work pages

  1. [1]

    Alfonsi, G. (2009). Reynolds-averaged Navier-Stokes equations for turbulence modeling . Applied Mechanics Reviews , 62(4):1--20

    work page 2009

  2. [2]

    Antonova, R., Rai, A., Li, T., and Kragic, D. (2020). Bayesian optimization in variational latent spaces with dynamic compression. In Conference on Robot Learning , pages 456--465. PMLR

    work page 2020

  3. [3]

    A., Wood, S

    Balan, A., Park, M. A., Wood, S. L., Anderson, W. K., Rangarajan, A., Sanjaya, D. P., and May, G. (2022). A review and comparison of error estimators for anisotropic mesh adaptation for flow simulations . Computers and Fluids , 234:105259

    work page 2022

  4. [4]

    C., Wilde, D., Kliemank, M., Reith, D., Foysi, H., and Kr \"a mer, A

    Bedrunka, M. C., Wilde, D., Kliemank, M., Reith, D., Foysi, H., and Kr \"a mer, A. (2021). Lettuce: Pytorch-based lattice boltzmann framework. In International Conference on High Performance Computing , pages 40--55. Springer

    work page 2021

  5. [5]

    Bellman, R. (1966). Dynamic programming. Science , 153(3731):34--37

    work page 1966

  6. [6]

    COIL: Constrained Optimization in Learned Latent Space: Learning Representations for Valid Solutions

    Bentley, P. J., Lim, S. L., Gaier, A., and Tran, L. (2022). Coil: Constrained optimization in learned latent space--learning representations for valid solutions. arXiv preprint arXiv:2202.02163

    work page Pith review arXiv 2022

  7. [7]

    Beume, N., Naujoks, B., and Emmerich, M. (2007). Sms-emoa: Multiobjective selection based on dominated hypervolume. European Journal of Operational Research , 181(3):1653--1669

    work page 2007

  8. [8]

    nearest neighbor

    Beyer, K., Goldstein, J., Ramakrishnan, R., and Shaft, U. (1999). When is “nearest neighbor” meaningful? In International conference on database theory , pages 217--235. Springer

    work page 1999

Show all 79 references
  1. [9]

    P., Higgins, I., Pal, A., Matthey, L., Watters, N., Desjardins, G., and Lerchner, A

    Burgess, C. P., Higgins, I., Pal, A., Matthey, L., Watters, N., Desjardins, G., and Lerchner, A. (2017). Understanding disentangling in Beta - VAE . In NIPS Workshop on Learning Disentangled Representations

    2017

  2. [10]

    A., Hu, X., and Thuerey, N

    Chen, L.-W., Cakal, B. A., Hu, X., and Thuerey, N. (2021). Numerical investigation of minimum drag profiles in laminar flow using deep learning surrogates. Journal of Fluid Mechanics , 919

    2021

  3. [11]

    Clune, J., Chen, A., and Lipson, H. (2013). Upload any object and evolve it: Injecting complex geometric patterns into cppns for further evolution. In 2013 IEEE Congress on Evolutionary Computation , pages 3395--3402. IEEE

    2013

  4. [12]

    O., Pennock, R

    Clune, J., Stanley, K. O., Pennock, R. T., and Ofria, C. (2011). On the performance of indirect encoding across the continuum of regularity. IEEE Transactions on Evolutionary Computation , 15(3):346--367

    2011

  5. [13]

    Collins, J., Cottier, B., and Howard, D. (2019). Comparing direct and indirect representations for environment-specific robot component design. In 2019 IEEE Congress on Evolutionary Computation (CEC) , pages 2705--2712. IEEE

    2019

  6. [14]

    Cully, A. (2019). Autonomous skill discovery with quality-diversity and unsupervised descriptors. In Proceedings of GECCO

    2019

  7. [15]

    Cully, A., Clune, J., Tarapore, D., and Mouret, J.-B. (2015). Robots that can adapt like animals. Nature , 521(7553):503--507

    2015

  8. [16]

    Deb, K., Pratap, A., Agarwal, S., and Meyarivan, T. (2002). A fast and elitist multiobjective genetic algorithm: Nsga-ii. IEEE transactions on evolutionary computation , 6(2):182--197

    2002

  9. [17]

    DeJong, K. (1975). Analysis of the behavior of a class of genetic adaptive systems. Dept. Computer and Communication Sciences, University of Michigan, Ann Arbor

    1975

  10. [18]

    and Nikolaidis, S

    Fontaine, M. and Nikolaidis, S. (2021). Differentiable quality diversity. Advances in Neural Information Processing Systems , 34

    2021

  11. [19]

    C., Togelius, J., Nikolaidis, S., and Hoover, A

    Fontaine, M. C., Togelius, J., Nikolaidis, S., and Hoover, A. K. (2020). Covariance matrix adaptation for the rapid illumination of behavior space. In Proceedings of the 2020 genetic and evolutionary computation conference , pages 94--102

    2020

  12. [20]

    I., S \'o bester, A., and Keane, A

    Forrester, A. I., S \'o bester, A., and Keane, A. J. (2007). Multi-fidelity optimization via surrogate modelling. Proceedings of the royal society a: mathematical, physical and engineering sciences , 463(2088):3251--3269

    2007

  13. [21]

    Gaier, A. (2015). Evolutionary design via indirect encoding of non-uniform rational basis splines. In Proceedings of the Companion Publication of the 2015 Annual Conference on Genetic and Evolutionary Computation , pages 1197--1200

    2015

  14. [22]

    Gaier, A., Asteroth, A., and Mouret, J.-B. (2018a). Data-efficient design exploration through surrogate-assisted illumination. Evolutionary computation , 26(3):381--410

    2018

  15. [23]

    Gaier, A., Asteroth, A., and Mouret, J.-B. (2018b). Data-efficient neuroevolution with kernel-based surrogate models. In Proceedings of the genetic and evolutionary computation conference , pages 85--92

    2018

  16. [24]

    Gaier, A., Asteroth, A., and Mouret, J.-B. (2019). Are quality diversity algorithms better at generating stepping stones than objective-based search? In Proceedings of the Genetic and Evolutionary Computation Conference Companion , pages 115--116

    2019

  17. [25]

    Gaier, A., Asteroth, A., and Mouret, J.-B. (2020). Discovering representations for black-box optimization. In Proceedings of the 2020 Genetic and Evolutionary Computation Conference , pages 103--111

    2020

  18. [26]

    N., Duvenaud, D., Hern \'a ndez-Lobato, J

    G \'o mez-Bombarelli, R., Wei, J. N., Duvenaud, D., Hern \'a ndez-Lobato, J. M., S \'a nchez-Lengeling, B., Sheberla, D., Aguilera-Iparraguirre, J., Hirzel, T. D., Adams, R. P., and Aspuru-Guzik, A. (2018). Automatic chemical design using a data-driven continuous representatio...

    2018

  19. [27]

    Hagg, A. (2020). Phenotypic Niching using Quality Diversity Algorithms . pages 1--30

    2020

  20. [28]

    Hagg, A. (2021). Discovering the preference hypervolume: an interactive model for real world computational co-creativity . PhD thesis, Leiden University

    2021

  21. [29]

    Hagg, A., Asteroth, A., and B \"a ck, T. (2020a). A deep dive into exploring the preference hypervolume. In International Conference on Computational Creativity , pages 394--397

    2020

  22. [30]

    Hagg, A., Berns, S., Asteroth, A., Colton, S., and B \"a ck, T. (2021). Expressivity of parameterized and data-driven representations in quality diversity search. In Proceedings of the Genetic and Evolutionary Computation Conference , pages 678--686

    2021

  23. [31]

    Hagg, A., Preuss, M., Asteroth, A., and B \"a ck, T. (2020b). An analysis of phenotypic diversity in multi-solution optimization. In International Conference on Bioinspired Methods and Their Applications , pages 43--55. Springer, Cham

    2020

  24. [32]

    Hagg, A., Wilde, D., Asteroth, A., and B \"a ck, T. (2020c). Designing air flow with surrogate-assisted phenotypic niching. In International Conference on Parallel Problem Solving from Nature , pages 140--153. Springer, Cham

    2020

  25. [33]

    Hagg, A., Zaefferer, M., Stork, J., and Gaier, A. (2019). Prediction of neural network performance by phenotypic modeling. In Proceedings of the Genetic and Evolutionary Computation Conference Companion , pages 1576--1582

    2019

  26. [34]

    M., Murman, E

    Hicks, R. M., Murman, E. M., and Vanderplaats, G. N. (1974). An assessment of airfoil design by numerical optimization. Technical report

    1974

  27. [35]

    and Branke, J

    Hildebrandt, T. and Branke, J. (2015). On using surrogates with genetic programming. Evolutionary computation , 23(3):343--367

    2015

  28. [36]

    Holland, J. H. (1975). Adaptation in natural and artificial systems . MIT press

    1975

  29. [37]

    Holland, J. H. (1992). Adaptation in natural and artificial systems: an introductory analysis with applications to biology, control, and artificial intelligence . MIT press

    1992

  30. [38]

    Hotz, P. E. (2004). Comparing direct and developmental encoding schemes in artificial evolution: A case study in evolving lens shapes. In Proceedings of the 2004 Congress on Evolutionary Computation (IEEE Cat. No. 04TH8753) , volume 1, pages 752--757. IEEE

    2004

  31. [39]

    M., Causon, D

    Ingram, D. M., Causon, D. M., and Mingham, C. G. (2003). Developments in Cartesian cut cell methods . Math. Comput. Simul. , 61(3-6):561--572

    2003

  32. [40]

    N., Ward, K

    Jacobs, E. N., Ward, K. E., and Pinkerton, R. M. (1933). The Characteristics of 78 related airfoil section from tests in the Variable-Density Wind Tunnel . Number 460. US Government Printing Office

    1933

  33. [41]

    Jin, Y. (2011). Surrogate-assisted evolutionary computation: Recent advances and future challenges. Swarm and Evolutionary Computation , 1(2):61--70

    2011

  34. [42]

    and Branke, J

    Kent, P. and Branke, J. (2020). Bop-elites, a bayesian optimisation algorithm for quality-diversity search. arXiv preprint arXiv:2005.04320

    2020 arXiv

  35. [43]

    Kicinger, R. (2006). Evolutionary developmental system for structural design. In AAAI Fall Symposium: Developmental Systems , pages 1--8

    2006

  36. [44]

    Kingma, D. P. and Welling, M. (2014). Auto- Encoding Variational Bayes . In Proceedings of ICLR

    2014

  37. [45]

    S., and Bandler, J

    Koziel, S., Cheng, Q. S., and Bandler, J. W. (2008). Space mapping. IEEE Microwave Magazine , 9(6):105--122

    2008

  38. [46]

    Lapok, P. (2020). Evolving planar mechanisms for the conceptual stage of mechanical design . PhD thesis, Edinburgh Napier University

    2020

  39. [47]

    L \" o hner, R. (2019). Towards overcoming the LES crisis . Int. J. Comut. Fluid Dyn. , 33(3):87--97

    2019

  40. [48]

    O., Mishra, S., and Ray, D

    Lye, K. O., Mishra, S., and Ray, D. (2020). Deep learning observables in computational fluid dynamics. Journal of Computational Physics , 410:109339

    2020

  41. [49]

    Mathieu, E., Rainforth, T., Siddharth, N., and Teh, Y. W. (2019). Disentangling disentanglement in variational autoencoders. In International Conference on Machine Learning , pages 4402--4412. PMLR

    2019

  42. [50]

    Mavriplis, D. J. (1997). Unstructured grid techniques . Annu. Rev. Fluid Mech. , 29:473--514

    1997

  43. [51]

    Wind comfort and wind danger in the built environment (in dutch)

    NEN 8100 (2006). Wind comfort and wind danger in the built environment (in dutch). Norm NEN 8100

    2006

  44. [52]

    O., and Glette, K

    Nordmoen, J., Veenstra, F., Ellefsen, K. O., and Glette, K. (2021). Map-elites enables powerful stepping stones and diversity for modular robotics. Frontiers in Robotics and AI , 8

    2021

  45. [53]

    Olhofer, M., Sendhoff, B., Arima, T., and Sonoda, T. (2000). Optimisation of a stator blade used in a transonic compressor cascade with evolution strategies. In Evolutionary Design and Manufacture , pages 45--54. Springer

    2000

  46. [54]

    and Huyer, W

    Po s \' k, P. and Huyer, W. (2012). Restarted local search algorithms for continuous black box optimization. Evolutionary Computation , 20(4):575--607

    2012

  47. [55]

    Preuss, M. (2012). Improved topological niching for real-valued global optimization. In European Conference on the Applications of Evolutionary Computation , pages 386--395. Springer

    2012

  48. [56]

    Rasmussen, C. E. and Nickisch, H. (2010). Gaussian processes for machine learning (gpml) toolbox. The Journal of Machine Learning Research , 11:3011--3015

    2010

  49. [57]

    Rios, T., Van Stein, B., B \"a ck, T., Sendhoff, B., and Menzel, S. (2021a). Point2ffd: Learning shape representations of simulation-ready 3d models for engineering design optimization. In 2021 International Conference on 3D Vision (3DV) , pages 1024--1033. IEEE

    2021

  50. [58]

    Rios, T., van Stein, B., Wollstadt, P., B \"a ck, T., Sendhoff, B., and Menzel, S. (2021b). Exploiting local geometric features in vehicle design optimization with 3d point cloud autoencoders. In 2021 IEEE Congress on Evolutionary Computation (CEC) , pages 514--521. IEEE

    2021

  51. [59]

    Sagaut, P. (2006). Large eddy simulation for incompressible flows: an introduction . Springer Science & Business Media

    2006

  52. [60]

    Computer as a Tool

    Sarakinos, S. S., Amoiralis, E., and Nikolos, I. K. (2005). Exploring freeform deformation capabilities in aerodynamic shape parameterization. In EUROCON 2005-The International Conference on" Computer as a Tool" , volume 1, pages 535--538. IEEE

    2005

  53. [61]

    Sederberg, T. W. and Parry, S. R. (1986). Free-form deformation of solid geometric models. ACM SIGGRAPH computer graphics , 20(4):151--160

    1986

  54. [62]

    Sobol', I. M. (1967). On the distribution of points in a cube and the approximate evaluation of integrals. Zhurnal Vychislitel'noi Matematiki i Matematicheskoi Fiziki , 7(4):784--802

    1967

  55. [63]

    Stanley, K. O. (2007). Compositional pattern producing networks: A novel abstraction of development. Genetic programming and evolvable machines , 8(2):131--162

    2007

  56. [64]

    O., D'Ambrosio, D

    Stanley, K. O., D'Ambrosio, D. B., and Gauci, J. (2009). A hypercube-based encoding for evolving large-scale neural networks. Artificial life , 15(2):185--212

    2009

  57. [65]

    Stanley, K. O. and Miikkulainen, R. (2002). Evolving neural networks through augmenting topologies. Evolutionary computation , 10(2):99--127

    2002

  58. [66]

    Stork, J., Zaefferer, M., and Bartz-Beielstein, T. (2019). Improving neuroevolution efficiency by surrogate model-based optimization with phenotypic distance kernels. In International Conference on the Applications of Evolutionary Computation (Part of EvoStar) , pages 504--519...

    2019

  59. [67]

    Stork, J., Zaefferer, M., Fischbach, A., Rehbach, F., and Bartz-Beielstein, T. (2017). Surrogate-assisted learning of neural networks

    2017

  60. [68]

    Sun, L., Gao, H., Pan, S., and Wang, J.-X. (2020). Surrogate modeling for fluid flows based on physics-constrained deep learning without simulation data. Computer Methods in Applied Mechanics and Engineering , 361:112732

    2020

  61. [69]

    F., and Yang, Y

    Tai, K., Wang, N. F., and Yang, Y. (2008). Target geometry matching problem with conflicting objectives for multiobjective topology design optimization using ga. In 2008 IEEE Congress on Evolutionary Computation (IEEE World Congress on Computational Intelligence) , pages 1873-...

    2008

  62. [70]

    Tripp, A., Daxberger, E., and Hern \'a ndez-Lobato, J. M. (2020). Sample-efficient optimization in the latent space of deep generative models via weighted retraining. Advances in Neural Information Processing Systems , 33

    2020

  63. [71]

    and Quagliarella, D

    Vicini, A. and Quagliarella, D. (1999). Airfoil and wing design through hybrid optimization strategies. AIAA journal , 37(5):634--641

    1999

  64. [72]

    Wales, D. J. and Doye, J. P. (1997). Global optimization by basin-hopping and the lowest energy structures of lennard-jones clusters containing up to 110 atoms. The Journal of Physical Chemistry A , 101(28):5111--5116

    1997

  65. [73]

    Wang, H., Jin, Y., and Yao, X. (2016). Diversity assessment in many-objective optimization. IEEE transactions on cybernetics , 47(6):1510--1522

    2016

  66. [74]

    Wang, Y., Shimada, K., and Farimani, A. B. (2021). Airfoil gan: Encoding and synthesizing airfoils for aerodynamic-aware shape optimization. arXiv preprint arXiv:2101.04757

    2021 arXiv

  67. [75]

    and Preuss, M

    Wessing, S. and Preuss, M. (2017). The true destination of ego is multi-local optimization. In 2017 IEEE Latin American Conference on Computational Intelligence (LA-CCI) , pages 1--6. IEEE

    2017

  68. [76]

    Wu, H., Liu, X., An, W., Chen, S., and Lyu, H. (2020). A deep learning approach for efficiently and accurately evaluating the flow field of supercritical airfoils. Computers & Fluids , 198:104393

    2020

  69. [77]

    Yannou, B., Dihlmann, M., and Cluzel, F. (2008). Indirect encoding of the genes of a closed curve for interactively create innovative car silhouettes. In 10th International Design Conference-DESIGN 2008 , pages 1243--1254

    2008

  70. [78]

    C., Hoover, A

    Zhang, Y., Fontaine, M. C., Hoover, A. K., and Nikolaidis, S. (2021). Deep surrogate assisted map-elites for automated hearthstone deckbuilding. arXiv e-prints , pages arXiv--2112

    2021

  71. [79]

    Zitzler, E., Laumanns, M., and Thiele, L. (2001). Spea2: Improving the strength pareto evolutionary algorithm. TIK-report , 103

    2001

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.