Pith. sign in

REVIEW 3 major objections 5 minor 44 references

Three-level surrogate stack cuts lattice optimization cost 24%

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · glm-5.2

2026-07-09 15:05 UTC pith:3EDE4ZKQ

load-bearing objection Useful multi-fidelity GA-tuning framework with a real but modest headline gap: the 7.158 GPa figure is CNN-predicted, not FFT-validated; the only FFT-validated config achieves 6.982 GPa (~2% below benchmark). the 3 major comments →

arxiv 2607.07289 v1 pith:3EDE4ZKQ submitted 2026-07-08 cond-mat.mtrl-sci cond-mat.dis-nncs.AIcs.LGmath.OC

Bayesian Optimization of Genetic Algorithm Hyperparameters in a Multi-Fidelity Framework for Efficient Lattice Material Design

Sergei Zorkaltsev , Maciej Haranczyk , Christina Schenk This is my paper
classification cond-mat.mtrl-sci cond-mat.dis-nncs.AIcs.LGmath.OC
keywords Bayesian optimizationgenetic algorithmmulti-fidelity optimizationlattice materialshyperparameter tuningconvolutional neural networksurrogate modelFFT homogenization
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper argues that the hyperparameters controlling a genetic algorithm (GA) used to design lattice materials can themselves be optimized automatically and cheaply by stacking three models of decreasing fidelity: a fast Gaussian process that searches the hyperparameter space, a 3D convolutional neural network that evaluates candidate lattice structures during each GA run, and high-fidelity FFT simulations reserved for final validation. The central object carrying the argument is this nested surrogate hierarchy: each layer accelerates the one above it, so that the expensive physics simulation is almost never called during the search itself. The paper shows that Bayesian optimization over GA hyperparameters, using the logNEI acquisition function to handle the noise inherent in stochastic GA runs, finds configurations allowing a 25-generation GA to match the elastic modulus achieved by a 75-generation baseline (7.158 GPa vs 7.119 GPa), cutting total compute from 225 to 171 hours. A penalized objective variant further reduces the number of lattice structures that must be evaluated, which matters when each structure would need to be physically fabricated and tested.

Core claim

The paper discovers that automated Bayesian optimization of GA hyperparameters, operating through a three-level multi-fidelity surrogate hierarchy, can recover the performance of a three-times-longer optimization at 24% lower computational cost, and that a penalized objective trades a small amount of performance for a large reduction in the number of structures requiring evaluation.

What carries the argument

The mechanism is a cascade of three fidelities: (1) a Gaussian process surrogate models the mapping from GA hyperparameters (number of parents, mutation fraction, cell mutation fraction) to achieved elastic modulus, guiding the hyperparameter search; (2) a DenseNet-based 3D CNN surrogate evaluates lattice structures during GA runs, replacing FFT simulations; (3) FFT-based homogenization validates final results. The logNEI acquisition function handles noisy GA evaluations by integrating over posterior uncertainty in both the latent objective and observations. A penalized objective divides the achieved modulus by a function of the parent population size, shifting optima toward configurations需要

Load-bearing premise

The CNN surrogate model, with a mean absolute error of 0.063 GPa, is assumed to be accurate enough that the hyperparameters identified as optimal when the GA uses the CNN are also near-optimal when the GA uses the true FFT evaluator. The paper validates only the final hyperparameter configurations with FFT, not the intermediate BO trajectory, so the surrogate's fidelity throughout the search space is taken on trust.

What would settle it

If the CNN surrogate systematically misranks lattice structures in some region of the design space, the GA runs guided by the CNN could converge to different optima than GA runs guided by FFT, meaning the BO-optimized hyperparameters would be optimal for the wrong objective function and would fail to transfer when validated with high-fidelity simulations.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the multi-fidelity cascade works for lattice stiffness optimization, the same architecture could tune GA hyperparameters for other discrete combinatorial design problems where each evaluation is expensive, such as truss topology or metamaterial unit-cell selection.
  • The finding that mutation disappears from optimal configurations suggests that crossover alone suffices to recombine high-performing structural building blocks in this design space, which could simplify future GA implementations for similar lattice problems.
  • The penalized objective formulation provides a direct bridge between computational optimization and experimental workflows, where fabrication cost per structure dominates and reducing the number of required specimens is the primary practical constraint.
  • The 24% cost reduction with preserved performance implies that the previous non-optimized hyperparameters were substantially sub-optimal, raising the question of how much further improvement remains accessible with additional BO iterations or higher-dimensional hyperparameter spaces.

Where Pith is reading between the lines

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

  • The disappearance of mutation in optimized configurations may reflect a property of the specific design space (five unit-cell types in a 4×4×4 grid) rather than a general principle; design spaces with more unit-cell types or larger assemblies might still require mutation for effective exploration.
  • If the CNN surrogate's error landscape is non-uniform across the design space, the BO-identified hyperparameters could be optimal for the surrogate but subtly sub-optimal for the true FFT objective, a risk the paper mitigates only by validating final configurations rather than the full search trajectory.
  • The framework could be extended to jointly optimize hyperparameters and the lattice objective formulation itself, since the penalized objective results suggest that the choice of what to maximize is as important as how to maximize it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. This manuscript presents a three-level multi-fidelity framework for optimizing the hyperparameters of a genetic algorithm (GA) used in lattice material design. The framework integrates high-fidelity FFT homogenization for validation, a medium-fidelity 3D CNN surrogate for rapid property evaluation during the GA, and a low-fidelity Gaussian process (GP) surrogate within a Bayesian optimization (BO) loop to guide the hyperparameter search. The authors evaluate four acquisition functions (UCB, NEI, qNEI, logNEI), finding that logNEI performs best. They show that optimized hyperparameters allow a 25-generation GA run to achieve specific elastic modulus values comparable to a 75-generation benchmark. Additionally, they introduce a penalized objective to reduce the number of parent lattices required, validating the practicality of the approach for experimental applications. The work is well-motivated, and the provision of reproducible code is a notable strength.

Significance. The paper addresses a relevant problem in computational materials design: the computational cost of tuning hyperparameters for evolutionary algorithms. The multi-fidelity strategy is logically structured, and the application of BO to GA hyperparameters in this context is a sensible approach. The authors provide publicly available code and data (GitHub), which is a significant strength that enhances reproducibility. The practical consideration of penalizing the number of parents to reduce experimental fabrication costs is a valuable contribution for the community. However, the significance of the central claim is somewhat tempered by the validation strategy, as detailed in the major comments.

major comments (3)
  1. The central claim that the framework achieves performance comparable to the 75-generation benchmark while reducing computational cost by 24% is undermined by a validation gap. The headline value of 7.158 GPa (Table 1, logNEI) is a CNN surrogate prediction, not an FFT-validated result. Section 3.1 explicitly states that during BO, all GA evaluations were carried out using the CNN surrogate. The only FFT validation performed (Section 3.3, Figure 5) uses the penalized objective configuration (α=0.20), which achieves 6.982 GPa—a 1.92% decrease relative to the 7.119 GPa benchmark. The unpenalized logNEI-optimal configuration (n_par=165, f_mut=0.646, f_cell=0.044) that produced the 7.158 GPa figure was never validated with FFT. The abstract's claim of 'preserving mechanical performance' rests on an unvalidated surrogate prediction, while the available ground-truth evidence shows a measurable,
  2. There is a mild circularity in the benchmarking setup. The CNN surrogate used during BO is retrained on data from the complete 75-generation optimization reported in Zorkaltsev et al. (2026)—the same study whose 75-generation result (7.119 GPa) serves as the benchmark the 25-generation optimized run is compared against. Thus, the 25-generation run benefits from information embedded in the surrogate that was expensive to generate, and this cost is excluded from the 24% reduction figure. The authors should explicitly acknowledge this dependency and clarify that the 24% reduction applies only to the online optimization cost, not the total cost of building the surrogate.
  3. The comparison of acquisition functions in Section 3.2 relies on a single BO trajectory per acquisition function. The authors acknowledge this limitation, but it is load-bearing for the claim that logNEI is superior. Given the inherent stochasticity of GA evaluations, a single trajectory is insufficient to establish a statistically significant ranking of acquisition functions. The authors should either soften the claim regarding logNEI's superiority or provide evidence (e.g., variance across multiple seeds) that the observed differences are robust.
minor comments (5)
  1. Section 2.3: The text describing FFT simulations is duplicated verbatim (the paragraph starting 'The high-fidelity evaluations in this work were performed using a spectral FFT-based computational solver...' appears twice). This should be corrected.
  2. Table 1: The column 'E_max, GPa' is labeled 'E_max, GPa' in the header but the values are presented without units in the rows. Ensure consistency.
  3. Section 3.3, Eq. (6): The penalized objective uses n_par with a tilde (ñ_par) in the equation, but the text refers to n_par. Clarify the notation to avoid confusion.
  4. Figure 5: The y-axis label 'Specific elastic modulus' should include units (GPa) for consistency with the text and tables.
  5. Section 2.5.1: The Matérn 5/2 kernel is defined with r_ij, but the ARD-scaled distance is written with a sum over k. Ensure the notation is clear and consistent, particularly regarding the placement of the lengthscale ℓ_k.

Circularity Check

0 steps flagged

No strict circularity; mild self-citation dependency and surrogate-training-data overlap, but the central claims are not forced by construction.

full rationale

The paper's derivation chain does not exhibit definitional circularity. The three fidelity levels (FFT, CNN, GP) are distinct models with distinct inputs: the GP models the hyperparameter→performance landscape, the CNN models lattice-structure→modulus, and FFT provides ground-truth mechanics. No equation reduces to its inputs by construction. The main concern raised by the skeptic — that the CNN surrogate was trained on data from the 75-generation run (Zorkaltsev et al. 2026) whose endpoint (7.119 GPa) serves as the benchmark for the 25-generation result (7.158 GPa) — is a methodological information-leakage concern, not a mathematical circularity. The CNN is a general regression model (lattice→property), not a memorization of the GA trajectory; the 25-generation GA still searches the design space and evaluates new candidate configurations through the CNN. The 7.158 GPa value is a CNN-predicted quantity that was not FFT-validated (only the penalized α=0.20 configuration was validated, achieving 6.982 GPa), but this is an overclaiming/correctness issue rather than a circularity where the prediction equals the fit by construction. The self-citation to Zorkaltsev et al. (2026) is load-bearing (provides CNN architecture, training data, benchmark, and GA framework), and the author sets overlap, but the cited work is a published peer-reviewed paper with its own independent validation, so the citation constitutes external evidence rather than an unverified self-referential chain. The 24% cost reduction excludes CNN training costs, which is a cost-accounting concern but not circularity. Overall, the framework's claims have independent content that is not tautologically forced by its inputs.

Axiom & Free-Parameter Ledger

7 free parameters · 5 axioms · 0 invented entities

The paper introduces no new physical entities, particles, or forces. The free parameters are the GA hyperparameters being optimized and the BO/GP configuration parameters. The axioms are standard domain assumptions for surrogate-assisted optimization, with two ad-hoc choices (Sobol point count, q value) that lack sensitivity analysis.

free parameters (7)
  • n_par = [10, 175]
    Number of parents selected per generation; one of three GA hyperparameters optimized by BO.
  • f_mut = [0.0, 1.0]
    Fraction of offspring subjected to mutation; GA hyperparameter optimized by BO.
  • f_cell = [0.0, 0.75]
    Fraction of cells modified within each mutated offspring; GA hyperparameter optimized by BO.
  • α (penalization weight) = 0.10, 0.15, 0.20, 0.25
    Manually chosen penalization parameter controlling the trade-off between elastic modulus and parent population size in the penalized objective (Eq. 6).
  • λ (UCB exploration parameter) = 1.0
    Exploration-exploitation balance for UCB acquisition function; set to β=1.0 without stated justification.
  • GP kernel lengthscales ℓ_k = fitted via Gamma priors
    ARD lengthscale parameters for the anisotropic Matérn 5/2 kernel, estimated during BO with Gamma priors.
  • σ²_f (signal variance) = fitted via Gamma prior
    GP signal variance, estimated during BO.
axioms (5)
  • domain assumption The anisotropic Matérn 5/2 kernel is an appropriate covariance function for modeling the GA hyperparameter-response landscape.
    Stated in §2.5.1; chosen because it 'permits moderately irregular behavior while retaining sufficient smoothness.' No comparison with alternative kernels is performed.
  • domain assumption The CNN surrogate (DenseNet, MAE 0.063 GPa) is sufficiently accurate to replace FFT simulations during the BO hyperparameter search.
    Stated in §3.1: 'all GA evaluations were carried out using the CNN surrogate, while full FFT-assisted GA runs were reserved exclusively for validation.' The accuracy of the CNN on the specific hyperparameter-driven configurations is not separately verified.
  • domain assumption The specific elastic modulus in the z-direction is an adequate single objective for lattice optimization.
    Used throughout; no multi-objective formulation or consideration of anisotropy, shear, or buckling is included.
  • ad hoc to paper 25 initial Sobol points provide sufficient coverage of the 3D hyperparameter space for BO initialization.
    Stated in §3.1; no sensitivity analysis to the number of initial points is provided.
  • ad hoc to paper q=3 repeated GA runs per BO iteration adequately capture the stochastic variability of GA evaluations.
    Mentioned in §2.5; the choice of q=3 is not justified by a variance analysis.

pith-pipeline@v1.1.0-glm · 16624 in / 3104 out tokens · 364745 ms · 2026-07-09T15:05:09.679081+00:00 · methodology

0 comments
read the original abstract

This study presents a multi-fidelity framework for the systematic optimization of genetic algorithm (GA) hyperparameters. The framework integrates three fidelity levels: high-fidelity Fast Fourier Transform (FFT) homogenization for validation, a medium-fidelity 3D convolutional neural network surrogate for rapid property evaluation, and a low-fidelity Gaussian process (GP) surrogate within a Bayesian optimization (BO) framework to guide the hyperparameter search. Various acquisition functions are evaluated, with logNEI achieving the best performance by effectively accounting for the noise inherent in GA evaluations. The proposed framework identifies hyperparameter configurations that enable a 25-generation GA run to achieve elastic modulus values comparable to those obtained in a full 75-generation optimization. Furthermore, introducing a penalized BO objective significantly reduces the number of required lattices with only minor decreases in absolute achieved elastic modulus, revealing a practical trade-off between performance and the number of structures that must be evaluated. High-fidelity FFT validation verifies the effectiveness of the surrogate-driven optimization strategy. The optimized hyperparameters allow for rapid convergence, eliminate the need for lattice mutation, and reduce the overall computational cost by 24% (from 225 to 171 hours) while preserving mechanical performance. These results demonstrate the potential of multi-fidelity optimization as an efficient and practical approach for GA hyperparameter tuning and future experimental lattice design studies.

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

44 extracted references · 44 canonical work pages · 2 internal anchors

  1. [1]

    Advanced Engineering Materials22(12), 2000611 (2020)

    Alomar, Z., Concli, F.: A review of the selective laser melting lattice structures and their numerical models. Advanced Engineering Materials22(12), 2000611 (2020)

    work page 2020

  2. [2]

    In: Oh, A., Naumann, T., Globerson, A., Saenko, K., Hardt, M., Levine, S

    Ament, S., Daulton, S., Eriksson, D., Balandat, M., Bakshy, E.: Unexpected improve- ments to expected improvement for bayesian optimization. In: Oh, A., Naumann, T., Globerson, A., Saenko, K., Hardt, M., Levine, S. (eds.) Advances in Neural Information Processing Systems, vol. 36, pp. 20577–20612. Curran Associates, Inc., Red Hook, NY (2023)

    work page 2023

  3. [3]

    Materialia38, 102233 (2024)

    Araya, M., Murillo, J., Vindas, R., Guill´ en, T.: Compressive behavior of SLA open-cell lattices: a comparison between triply periodic minimal surface gyroid and stochastic structures for artificial bone. Materialia38, 102233 (2024)

    work page 2024

  4. [4]

    In: Advances in Neural Information Processing Systems 33 (2020)

    Bakshy, E.: BoTorch: A Framework for Efficient Monte-Carlo Bayesian Opti- mization. In: Advances in Neural Information Processing Systems 33 (2020). https://proceedings.neurips.cc/paper/2020/hash/f5b1b89d98b7286673128a5fb112cb9a- Abstract.html

    work page 2020

  5. [5]

    Materials & Design208, 109937 (2021)

    Challapalli, A., Patel, D., Li, G.: Inverse machine learning framework for optimizing lightweight metamaterials. Materials & Design208, 109937 (2021)

    work page 2021

  6. [6]

    Mathematics of Computation19(90), 297–301 (1965) https://doi

    Cooley, J.W., Tukey, J.W.: An algorithm for the machine calculation of complex Fourier series. Mathematics of Computation19(90), 297–301 (1965) https://doi. org/10.1090/S0025-5718-1965-0178586-1 Dos Reis, F., Karathanasopoulos, N.: Inverse metamaterial design combining genetic algorithms with asymptotic homogenization schemes. International Journal of Sol...

    work page doi:10.1090/s0025-5718-1965-0178586-1 1965

  7. [7]

    International Journal of Steel Structures16, 743–753 (2016)

    Feng, R.-q., Liu, F.-c., Xu, W.-j., Ma, M., Liu, Y.: Topology optimization method of lattice structures based on a genetic algorithm. International Journal of Steel Structures16, 743–753 (2016)

    work page 2016

  8. [8]

    A Tutorial on Bayesian Optimization

    Frazier, P.I.: A tutorial on Bayesian optimization. arXiv preprint arXiv:1807.02811 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  9. [9]

    QFT-based Homogenization

    Givois, F., Kabel, M., Gauger, N.: QFT-based Homogenization (2022). https://arxiv. org/abs/2207.12949

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  10. [10]

    Procedia Cirp88, 411–416 (2020)

    Minetola, P., Fino, P., Iuliano, L.: Ti-6Al-4V lattice structures produced by EBM: 16 Heat treatment and mechanical properties. Procedia Cirp88, 411–416 (2020)

    work page 2020

  11. [11]

    Garland, A.P., White, B.C., Jensen, S.C., Boyce, B.L.: Pragmatic generative opti- mization of novel structural lattice metamaterials with machine learning. Materials & Design203, 109632 (2021) Hern´ andez-del-Valle, M., Schenk, C., Echevarr´ ıa-Pastrana, L., Ozdemir, B., Dios- L´ azaro, E., Ilarraza-Zuazo, J., Wang, D.-Y., Haranczyk, M.: Robotically autom...

    work page doi:10.1039/d3dd00141e 2021

  12. [12]

    Composite Structures323, 117454 (2023)

    Hosseini, S., Farrokhabadi, A., Chronopoulos, D.: Experimental and numerical anal- ysis of shape memory sinusoidal lattice structure: Optimization through fusing an artificial neural network to a genetic algorithm. Composite Structures323, 117454 (2023)

    work page 2023

  13. [13]

    In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp

    Huang, G., Liu, Z., Van Der Maaten, L., Weinberger, K.Q.: Densely connected con- volutional networks. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 4700–4708 (2017)

    work page 2017

  14. [14]

    Computer Methods in Applied Mechanics and Engineering388, 114223 (2022)

    Lucarini, S., Cobian, L., Voitus, A., Segurado, J.: Adaptation and validation of FFT methods for homogenization of lattice based materials. Computer Methods in Applied Mechanics and Engineering388, 114223 (2022)

    work page 2022

  15. [15]

    Journal of Materials Engineering and Performance33(10), 4685–4711 (2024)

    Liu, R., Chen, W., Zhao, J.: A review on factors affecting the mechanical properties of additively-manufactured lattice structures. Journal of Materials Engineering and Performance33(10), 4685–4711 (2024)

    work page 2024

  16. [16]

    Bayesian Analysis14(2), 495–519 (2019) https://doi.org/ 10.1214/18-BA1110

    Letham, B., Karrer, B., Ottoni, G., Bakshy, E.: Constrained Bayesian optimization with noisy experiments. Bayesian Analysis14(2), 495–519 (2019) https://doi.org/ 10.1214/18-BA1110

    work page doi:10.1214/18-ba1110 2019

  17. [17]

    Computational Mechanics63(2), 365–382 (2019)

    Lucarini, S., Segurado, J.: On the accuracy of spectral solvers for micromechanics based fatigue modeling. Computational Mechanics63(2), 365–382 (2019)

    work page 2019

  18. [18]

    Materials Horizons9(3), 952–960 (2022)

    Lee, S., Zhang, Z., Gu, G.X.: Generative machine learning algorithm for lattice structures with superior mechanical properties. Materials Horizons9(3), 952–960 (2022)

    work page 2022

  19. [19]

    Journal of Mechanical Design142(9), 091705 (2020)

    Liu, Y., Zhuo, S., Xiao, Y., Zheng, G., Dong, G., Zhao, Y.F.: Rapid modeling and design optimization of multi-topology lattice structure based on unit-cell library. Journal of Mechanical Design142(9), 091705 (2020)

    work page 2020

  20. [20]

    Computer Methods in Applied Mechanics and Engineering157(1), 69–94 (1998) https://doi.org/10.1016/ S0045-7825(97)00218-1 17

    Moulinec, H., Suquet, P.: A numerical method for computing the overall response of nonlinear composites with complex microstructure. Computer Methods in Applied Mechanics and Engineering157(1), 69–94 (1998) https://doi.org/10.1016/ S0045-7825(97)00218-1 17

    work page 1998

  21. [21]

    Scientific Reports10(1), 17663 (2020) https://doi.org/10.1038/s41598-020-74394-1

    Noack, M.M., Doerk, G.S., Li, R., Streit, J.K., Vaia, R.A., Yager, K.G., Fukuto, M.: Autonomous materials discovery driven by Gaussian process regression with inho- mogeneous measurement noise and anisotropic kernels. Scientific Reports10(1), 17663 (2020) https://doi.org/10.1038/s41598-020-74394-1 . Number: 1 Publisher: Nature Publishing Group. Accessed 2...

    work page doi:10.1038/s41598-020-74394-1 2020

  22. [22]

    APL Machine Learning2(1), 010902 (2024) https: //doi.org/10.1063/5.0176963

    Noack, M.M., Luo, H., Risser, M.D.: A unifying perspective on non-stationary kernels for deeper Gaussian processes. APL Machine Learning2(1), 010902 (2024) https: //doi.org/10.1063/5.0176963 . Accessed 2026-06-15

    work page doi:10.1063/5.0176963 2024

  23. [23]

    Advanced Intelligent Discovery, 202500054 https://doi.org/10.1002/aidi.202500054

    Ozdemir, B., Hern´ andez-del-Valle, M., Schenk, C., Wang, D.-Y., Haranczyk, M.: Bayesian optimization guiding the experimental mapping of the Pareto front of mechanical and flame-retardant properties in polyamide nanocomposites. Advanced Intelligent Discovery, 202500054 https://doi.org/10.1002/aidi.202500054

    work page doi:10.1002/aidi.202500054

  24. [24]

    Advances in neural information processing systems32(2019)

    Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al.: PyTorch: An imperative style, high-performance deep learning library. Advances in neural information processing systems32(2019)

    work page 2019

  25. [25]

    Materials14(12), 3225 (2021)

    Gianella, S., Barbato, M., Ortona, A.: Application of ceramic lattice structures to design compact, high temperature heat exchangers: material and architecture selection. Materials14(12), 3225 (2021)

    work page 2021

  26. [26]

    International Journal of Mechanical Sciences88, 154–161 (2014)

    Ravari, M.K., Kadkhodaei, M., Badrossamay, M., Rezaei, R.: Numerical investigation on mechanical properties of cellular lattice structures fabricated by fused deposition modeling. International Journal of Mechanical Sciences88, 154–161 (2014)

    work page 2014

  27. [27]

    Press, Cambridge, Massachusetts (2006)

    work page 2006

  28. [28]

    Materials14(6), 1366 (2021)

    Rahman, H., Yarali, E., Zolfagharian, A., Serjouei, A., Bodaghi, M.: Energy absorp- tion and mechanical performance of functionally graded soft–hard lattice structures. Materials14(6), 1366 (2021)

    work page 2021

  29. [29]

    Advanced Engineering Materials25(20), 2201385 (2023)

    Schwahofer, O., B¨ uttner, S., Binder, J., Colin, D., Drechsler, K.: Multiscale optimiza- tion of 3D-printed beam-based lattice structures through elastically tailored unit cells. Advanced Engineering Materials25(20), 2201385 (2023)

    work page 2023

  30. [30]

    Acta Mechanica232(6), 2051–2100 (2021) https://doi.org/10.1007/ s00707-021-02962-1

    Schneider, M.: A review of nonlinear FFT-based computational homogenization methods. Acta Mechanica232(6), 2051–2100 (2021) https://doi.org/10.1007/ s00707-021-02962-1

    work page 2051

  31. [31]

    Advanced Engineering Informatics 18 76, 104960 (2026) https://doi.org/10.1016/j.aei.2026.104960

    Schenk, C., Hern´ andez-del-Valle, M., Calero-Lumbreras, L., Noack, M., Haranczyk, M.: Noise-aware optimization in nominally identical manufacturing and measuring systems for high-throughput parallel workflows. Advanced Engineering Informatics 18 76, 104960 (2026) https://doi.org/10.1016/j.aei.2026.104960

    work page doi:10.1016/j.aei.2026.104960 2026

  32. [32]

    Composite Structures284, 115159 (2022)

    Shahrzadi, M., Emami, M.D., Akbarzadeh, A.: Heat transfer in BCC lattice materials: Conduction, convection, and radiation. Composite Structures284, 115159 (2022)

    work page 2022

  33. [33]

    Composite Structures283, 115102 (2022)

    Sharma, D., Hiremath, S.S.: Bio-inspired repeatable lattice structures for energy absorption: Experimental and finite element study. Composite Structures283, 115102 (2022)

    work page 2022

  34. [34]

    Results in Materials13, 100242 (2022)

    Spear, D.G., Lane, J.S., Palazotto, A.N., Kemnitz, R.A.: Computational based inves- tigation of lattice cell optimization under uniaxial compression load. Results in Materials13, 100242 (2022)

    work page 2022

  35. [35]

    Proceedings of the IEEE 104(1), 148–175 (2016) https://doi.org/10.1109/JPROC.2015.2494218

    Shahriari, B., Swersky, K., Wang, Z., Adams, R.P., Freitas, N.: Taking the Human Out of the Loop: A Review of Bayesian Optimization. Proceedings of the IEEE 104(1), 148–175 (2016) https://doi.org/10.1109/JPROC.2015.2494218 . Accessed 2025-05-20

    work page doi:10.1109/jproc.2015.2494218 2016

  36. [36]

    International Journal of Heat and Mass Transfer47(14-16), 3171–3186 (2004)

    Tian, J., Kim, T., Lu, T., Hodson, H., Queheillalt, D., Sypeck, D., Wadley, H.: The effects of topology upon fluid-flow and heat-transfer within cellular copper structures. International Journal of Heat and Mass Transfer47(14-16), 3171–3186 (2004)

    work page 2004

  37. [37]

    Mechanics of Materials, 105399 (2025) https://doi.org/10.1016/j.mechmat.2025.105399

    Wang, K., Gao, X.-L.: Inverse design of interpenetrating phase composites with targeted stiffness through deep learning. Mechanics of Materials, 105399 (2025) https://doi.org/10.1016/j.mechmat.2025.105399

    work page doi:10.1016/j.mechmat.2025.105399 2025

  38. [38]

    Materials14(16), 4410 (2021)

    Wen, Z., Li, M.: Compressive properties of functionally graded bionic bamboo lattice structures fabricated by fdm. Materials14(16), 4410 (2021)

    work page 2021

  39. [39]

    Additive Manufacturing60, 103238 (2022)

    Wang, J., Panesar, A.: Machine learning based lattice generation method derived from topology optimisation. Additive Manufacturing60, 103238 (2022)

    work page 2022

  40. [40]

    International Journal of Mechanical Sciences, 110514 (2025) https://doi.org/10.1016/j.ijmecsci.2025.110514

    Xiang, Y., Hou, J., Chen, X., Tang, K., Wang, X.: Decoupled design of hybrid mechan- ical metamaterials via ensembled deep learning. International Journal of Mechanical Sciences, 110514 (2025) https://doi.org/10.1016/j.ijmecsci.2025.110514

    work page doi:10.1016/j.ijmecsci.2025.110514 2025

  41. [41]

    International Journal of Mechanical Sciences219, 107093 (2022)

    Xiao, L., Xu, X., Feng, G., Li, S., Song, W., Jiang, Z.: Compressive performance and energy absorption of additively manufactured metallic hybrid lattice structures. International Journal of Mechanical Sciences219, 107093 (2022)

    work page 2022

  42. [42]

    Advanced Materials31(34), 1803670 (2019)

    Yeo, S.J., Oh, M.J., Yoo, P.J.: Structurally controlled cellular architectures for high-performance ultra-lightweight materials. Advanced Materials31(34), 1803670 (2019)

    work page 2019

  43. [43]

    Journal of Materials 19 Research and Technology30, 7523–7532 (2024)

    Zheng, Q., Chen, H., Zhou, J., Wang, W., Zheng, L., Xi, S.: Lightweight design of lat- tice structure of boron steel prepared by selective laser melting. Journal of Materials 19 Research and Technology30, 7523–7532 (2024)

    work page 2024

  44. [44]

    Computational Materials Science262, 114332 (2026) https://doi

    Zorkaltsev, S., Segurado, J., P´ erez-Prado, M.T., Haranczyk, M.: Does high entropy improve elastic properties of 3d lattice materials?—a genetic algorithm and active learning study. Computational Materials Science262, 114332 (2026) https://doi. org/10.1016/j.commatsci.2025.114332 20

    work page doi:10.1016/j.commatsci.2025.114332 2026