Pith. sign in

REVIEW 4 major objections 5 minor 50 references

Optimal design of frame structures with mixed categorical and continuous design variables using the Gumbel-Softmax method

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Frame-structure design with discrete catalog choices can be optimized by gradients, not just genetic algorithms, using Gumbel-Softmax sampling.

desk verdict Genuine first use of Gumbel-Softmax for frame optimization, with a real speedup in three case studies; the straight-through gradient is a heuristic without guarantees, but the paper is honest about that. read the letter →

arxiv 2501.00258 v1 pith:2Q4VI5YY submitted 2024-12-31 cs.CE math.OC

Mehran Ebrahimi , Hyunmin Cheong , Pradeep Kumar Jayaraman , Farhad Javid This is my paper · ORCID
classification cs.CEmath.OC MSC 74P1090C2690C30
keywords Gumbel-Softmaxcategoricaldesignvariablesframestructuresstructuraloptimizationgradient-basedadjointsensitivityanalysisgeneticalgorithmcomparisonfiniteelement
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that categorical design variables in frame structures—choices such as beam cross-section profiles from a standard catalog—can be handled by gradient-based optimization once each choice is reparameterized as a probability distribution and sampled differentiably. Its optimizer, GSMO, uses the Gumbel-Softmax relaxation so that each optimization iteration needs only one finite-element solve while updating both categorical and continuous variables. In three case studies, GSMO matched or beat a genetic algorithm on solution quality and ran about three orders of magnitude faster on the two large problems. If the argument holds, structural optimization with hundreds of discrete catalog choices no longer requires expensive population-based search.

What carries the argument

The central object is the straight-through Gumbel-Softmax estimator: it draws a differentiable soft sample $e_s = \operatorname{softmax}((\theta+G)/\tau)$, takes a hard sample $b_s$ by argmax for the forward simulation, and uses $\nabla_\theta e_s$ in place of the nondifferentiable $\nabla_\theta b_s$ during backpropagation. A per-variable attribute matrix $A_m$ maps soft sample entries to the continuous properties (area, moments of inertia, modulus) of each catalog choice, so one adjoint solve yields gradients of mass, compliance, stress, or frequency with respect to choice probabilities.

What would settle it

Run the optimizer on a small truss whose global optimum is known and record, over many random seeds, how often an update computed with $\nabla_\theta e_s$ reduces the true objective computed with $b_s$; if the estimated gradient frequently points uphill, the straight-through assumption fails. A complementary check is to compare $\nabla_\theta e_s$ against a finite-difference estimate of $\nabla_\theta \mathbb{E}[J(b_s)]$ at moderate temperature—large divergence would show the relaxation is not a faithful descent direction.

Watch

Extended reading notes

Core claim

The paper's central claim is that Gumbel-Softmax makes categorical structural variables amenable to sensitivity analysis, so a single gradient-based optimizer can solve mixed categorical-continuous frame design problems. Each categorical variable is reparameterized as unnormalized log-probabilities $\theta$; a hard one-hot sample $b_s$ is drawn for the finite-element solve, while the straight-through estimator substitutes the differentiable soft-sample gradient $\nabla_\theta e_s$ for the nondifferentiable hard-sample gradient. In the 72-bar truss, 812-bar lattice, and 258-bar bridge problems, the method found solutions comparable to or better than a genetic algorithm, with ten runs clustering closely around the best value, and it needed roughly 1000 times fewer finite-element solves than the GA baseline in the two large cases.

Load-bearing premise

Everything rests on the assumption that the gradient of a softened mixture of choices is a good stand-in for the gradient of the actual single choice used in the simulation.

Editorial extensions

If this is right

  • One finite-element solve per optimization iteration suffices, so per-iteration cost stops scaling with the number of design variables.
  • Categorical and continuous variables can be optimized simultaneously in one loop; in the 258-bar bridge case, simultaneous GSMO outperformed the bilevel BiGSMO variant.
  • On the tested problems, GSMO and BiGSMO were about 1000 times faster than the implemented genetic algorithm and produced more consistent solutions across runs.
  • The same algorithm transfers to any objective and constraint functions whose adjoint sensitivities are computable, including modal-analysis constraints as in the bridge case.

Reading between the lines

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

  • Editorial inference: because the straight-through gradient is a biased stand-in for the true discrete gradient, the paper's own report of oscillatory behavior on variables with many choices suggests a bias–variance trade-off; averaging several Gumbel samples per variable is a direct test of that diagnosis.
  • Editorial inference: the attribute-matrix requirement limits the method to catalogs whose choices share a continuous parameterization; purely symbolic choices such as joint types would need a learned embedding before the same gradient path applies.
  • Editorial inference: comparing GSMO against an unbiased gradient estimator on the same benchmarks would separate the benefit of gradient information from the benefit of the particular relaxation bias, and would clarify whether the speedup is intrinsic to gradient methods or specific to Gumbel-Softmax.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes replacing categorical design variables in frame-structure optimization by unnormalized log-probabilities and using the Gumbel-Softmax method with a straight-through estimator to obtain approximate gradients, while continuous variables are handled through adjoint sensitivity analysis. This yields two optimizers, GSMO and BiGSMO, which require one finite element solve per iteration regardless of the number of design variables. Three case studies compare the proposed methods with a genetic algorithm in terms of objective value, consistency over 10 runs, and wall-clock time. The sensitivity derivations and the Gumbel-max distribution proof in Appendix A are correct, and the measured runtime advantage over the implemented GA is substantial. However, the central claim that the straight-through soft gradient is a reliable descent direction for the hard discrete objective is not proven, and the GA comparisons are stopped well short of convergence in the larger cases; the conclusion also concedes failure modes in the high-cardinality regime that the paper otherwise advertises as its main advantage.

Significance. If the proposed heuristic is reliable, this is a significant contribution: it would allow gradient-based optimization of structural problems with hundreds of categorical choices such as cross-sectional profiles, reducing the number of finite element solves by orders of magnitude relative to population-based methods. The strengths of the paper are its clearly stated algorithm, correct adjoint formulas, explicit treatment of the sampling process, and honest reporting of limitations. The main weakness is that the load-bearing approximation--replacing the hard sample gradient with the soft Gumbel-Softmax gradient--is asserted to be accurate as the temperature anneals but is never analyzed or bounded. The empirical demonstrations are suggestive but not conclusive because the GA baseline is under-budgeted and the high-cardinality regime is not tested. The contribution is therefore best seen as a promising heuristic rather than a validated method with proven convergence.

major comments (4)
  1. [Section 4.2, Eq. (20); Algorithm 2] The central gradient identity mixes the hard forward sample b_s with the soft backward sample e_s: J and ∇_a J are evaluated at b_s while ∇_θ e_s is used in place of ∇_θ b_s. Since b_s is piecewise constant, ∇_θ b_s = 0 almost everywhere, so Eq. (20) is not the gradient of any single differentiable objective. The assertion that the discrepancy 'diminishes' as e_s approaches b_s is qualitative; no bias bound, variance estimate, or convergence result for the relaxed problem is given. This is load-bearing because the headline advantage rests on the reliability of this direction. I would like to see either a formal analysis (e.g., a bias bound in terms of τ and the number of choices, or a stationarity result for the annealed problem) or an explicit repositioning of GSMO/BiGSMO as heuristics, with the conclusion's overclaim about 'optimal solutions' softened.
  2. [Section 5.2 and Table 5] The empirical comparison does not close the gap left by the missing convergence analysis. GA is limited to 100 iterations in the bridge and lattice problems, yet the paper itself notes that GA became feasible only after about 300 iterations in the lattice problem. In the bridge problem, GA's average objective is 20.520 with standard deviation 2.712 versus a best of 13.642, which indicates GA is far from converged at the stopping point. The runtime advantage (one FE solve per iteration versus thousands per GA generation) is solid evidence of per-iteration cost, but it is not evidence that GSMO finds better optima than a converged GA. Please report GA results under a larger budget or include a convergence study that shows GA's objective has stagnated.
  3. [Section 6, Conclusion] The conclusion admits that as the number of choices per categorical variable increases, GSMO and BiGSMO 'exhibit oscillatory convergence behavior and, occasionally, may even fail to converge to a mathematically optimal solution.' This is exactly the regime claimed as the method's advantage, since the paper motivates the approach with 'hundreds of categorical choices.' The case studies use only 5 choices per categorical variable in the bridge and 4 in the lattice, so they do not demonstrate performance in the high-cardinality regime. The claim that the method scales to large numbers of choices should either be supported by experiments with larger choice sets or removed and qualified.
  4. [Section 6, first paragraph] The statement that the method 'transformed combinatorial optimization problems ... into problems involving only continuous design variables with a polynomial complexity' is an overstatement. The relaxed stochastic problem is still nonconvex, and a polynomial per-iteration cost does not imply polynomial total effort or global optimality. I suggest reporting the per-iteration complexity and the observed number of iterations instead of asserting polynomial complexity for the overall problem.
minor comments (5)
  1. [Section 4.2, Eq. (17)] Equation (17) uses conflicting row/column conventions: the first expression in (17) treats ∇_a J as a column vector, while the second expression and Eq. (20) treat it as a row vector. Please make the convention uniform.
  2. [Section 4.3, Algorithm 3, Step 18] Step 18 says ∇_{a_i} J is computed 'utilizing ... ∇_{θ_i} e_{s_i} found in Step 6 and employing (21)', but Eq. (21) does not involve ∇_{θ_i} e_{s_i}; the categorical gradient is assembled only in Steps 18-19 via Eq. (20). Please correct the cross-reference.
  3. [Section 5.2, Table 3] The sentence 'All GA runs could find a feasible solution after about 300 iterations' is in tension with Table 3, which reports GA optimum choices; clarify how GA's reported design was obtained and whether the 100-iteration runs had feasible designs.
  4. [Section 6, Conclusion] The conclusion's 'O(103) lower' should read O(10^3) or 'three orders of magnitude'; the typeset form is ambiguous.
  5. [Section 5.1, Table 2] For the 72-bar truss, Table 2 reports only the best GA solution; since the paper emphasizes consistency (mean and standard deviation over 10 runs), the corresponding GA mean and standard deviation should be reported for a fair comparison.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the Gumbel-Softmax gradient and adjoint sensitivities are derived from stated assumptions, and the case-study comparisons rest on external benchmarks rather than on fitted inputs.

full rationale

The paper's central claim—that GSMO/BiGSMO reduce computational cost by requiring only one finite-element solve per iteration—is an algorithmic property rather than a fitted result. The Gumbel-Softmax derivation is self-contained: Eq. (7) defines the soft sample, Eq. (10) gives its derivative, and Eq. (20) is explicitly presented as the straight-through approximation replacing the nondifferentiable hard-sample gradient with the soft-sample gradient; the asserted diminishing discrepancy is an unproved convergence assumption, which is a correctness risk but not circular. The attribute matrix A_m is an input describing the available cross-sections, and ∇aJ is computed by a standard adjoint method rather than calibrated to the reported optima. The 72-bar truss benchmark comes from external literature [20], and the genetic algorithm is an independent baseline; no case-study objective is used to fit hyperparameters in a way that forces the reported outcomes. The paper's self-citations support background tools such as beam elements, adjoint methods, and configuration search, but none is load-bearing for the claimed prediction. The conclusion's admission that many categorical choices may cause oscillatory or failed convergence is a stated limitation, not evidence that the derivation reduces to its own inputs.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central method rests on standard probability theory for Gumbel-Max sampling, a domain modeling assumption that catalog choices have continuous attributes, and an unproven heuristic that the straight-through gradient provides a valid descent direction. No new physical entities are introduced. Hyperparameters such as temperature schedule and step size are hand-set and could affect the reported results.

free parameters (3)
  • Gumbel temperature initial value and annealing schedule = tau_0 = 100, decay factor 0.9 per iteration, floor 0.01
    Hand-chosen hyperparameters that control relaxation bias and exploration/exploitation; no sensitivity study is reported (Section 5).
  • Optimization step size = 1e-3
    Fixed step size used for all design variables; no adaptive scheme or line search is described (Section 5).
  • Genetic algorithm comparison hyperparameters = population size = 10x design variables, crossover 0.9, mutation 0.1, penalty factor 1000, 100 iterations
    These settings affect the reported runtime and quality comparison with GSMO; they are reasonable but not tuned, so the speedup figures are implementation-dependent.
assumptions (4)
  • standard math Gumbel-Max samples generated by argmax of Gumbel-perturbed logits follow the categorical distribution associated with the logits.
    Standard result proven in Appendix A using the Gumbel cumulative distribution and density functions.
  • domain assumption Every categorical choice can be characterized by a common vector of continuous attributes, and the objective and constraint functions are differentiable with respect to those attributes.
    Required for Eq. (20); the paper itself acknowledges this fails for choices such as joint types, which have no obvious continuous attribute representation (Section 6).
  • ad hoc to paper The straight-through estimator's soft gradient is an acceptable substitute for the hard sample gradient and provides a useful descent direction.
    No proof or error bound is given; reliance is explicit in Eqs. (8)-(10), and the conclusion warns of possible non-convergence with many choices.
  • domain assumption The structural response is governed by linear elasticity with external forces independent of displacements.
    Used in Section 4.1 to derive the adjoint equation (14) and the sensitivity formula (15); the paper limits its claims to this regime.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Optimal design of frame structures with mixed categorical and continuous design variables using the Gumbel-Softmax method." pith.science (2026). https://pith.science/paper/2Q4VI5YY

@misc{pith2026250100258,
  author       = {Pith},
  title        = {Pith review of: Optimal design of frame structures with mixed categorical and continuous design variables using the Gumbel-Softmax method},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2Q4VI5YY}},
  note         = {Machine review of arXiv:2501.00258}
}
read the original abstract

In optimizing real-world structures, due to fabrication or budgetary restraints, the design variables may be restricted to a set of standard engineering choices. Such variables, commonly called categorical variables, are discrete and unordered in essence, precluding the utilization of gradient-based optimizers for the problems containing them. In this paper, incorporating the Gumbel-Softmax (GSM) method, we propose a new gradient-based optimizer for handling such variables in the optimal design of large-scale frame structures. The GSM method provides a means to draw differentiable samples from categorical distributions, thereby enabling sensitivity analysis for the variables generated from such distributions. The sensitivity information can greatly reduce the computational cost of traversing high-dimensional and discrete design spaces in comparison to employing gradient-free optimization methods. In addition, since the developed optimizer is gradient-based, it can naturally handle the simultaneous optimization of categorical and continuous design variables. Through three numerical case studies, different aspects of the proposed optimizer are studied and its advantages over population-based optimizers, specifically a genetic algorithm, are demonstrated.

Figures

Figures reproduced from arXiv: 2501.00258 by the authors.

Figure 1
Figure 1. The 72-bar truss structure. The typical story is repeated 4 times along the Z axis. The [PITH_FULL_IMAGE:figures/full_fig_p015_1.png] view at source ↗
Figure 2
Figure 2. Convergence plot of the 10 GSMO and GA runs for the truss structure [PITH_FULL_IMAGE:figures/full_fig_p017_2.png] view at source ↗
Figure 3
Figure 3. Geometry of the lattice structure 20 mm 10 mm 20 mm 10 mm 5 mm 10 mm CS1 CS2 CS3 CS4 [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Available cross-sectional choices for the beams in the lattice structure [PITH_FULL_IMAGE:figures/full_fig_p018_4.png]
Figure 5
Figure 5. Figure 5: Optimum solutions of the lattice structure due to GSMO and BiGSMO [PITH_FULL_IMAGE:figures/full_fig_p019_5.png]
Figure 6
Figure 6. Figure 6: The 258-bar bridge structure The objective function of this problem is to minimize the total strain energy of the structure due to the applied loads while the maximum stress in all elements remains below their yield strength. Also, the smallest natural frequency of the…
Figure 7
Figure 7. Figure 7: Available cross-sectional choices for the beams in the bridge structure [PITH_FULL_IMAGE:figures/full_fig_p021_7.png]
Figure 8
Figure 8. Figure 8: Convergence plot of GSMO, BiGSMO and GA averaged over the 10 runs for the bridge [PITH_FULL_IMAGE:figures/full_fig_p022_8.png]
Figure 9
Figure 9. Figure 9: Optimum solution of the bridge structure due to GSMO. The deformations are scaled by [PITH_FULL_IMAGE:figures/full_fig_p024_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 49 canonical work pages

  1. [1]

    McGuire, R

    W. McGuire, R. H. Gallagher, and H. Saunders, Matrix structural analysis . John Wiley & Sons, 1982

    work page 1982

  2. [2]

    Optimal design of steel structures using standard sections,

    M.-W. Huang and J. S. Arora, “Optimal design of steel structures using standard sections,” Structural Optimization, vol. 14, no. 1, pp. 24–35, 1997

    work page 1997

  3. [3]

    Charged system search for optimal design of frame structures,

    A. Kaveh and S. Talatahari, “Charged system search for optimal design of frame structures,” Applied Soft Computing , vol. 12, no. 1, pp. 382–393, 2012

    work page 2012

  4. [4]

    A low order, torsion deformable spatial beam element based on the absolute nodal coordinate formulation and bishop frame,

    M. Ebrahimi, A. Butscher, and H. Cheong, “A low order, torsion deformable spatial beam element based on the absolute nodal coordinate formulation and bishop frame,” Multibody System Dynamics , vol. 51, no. 3, pp. 247–278, 2021

    work page 2021

  5. [5]

    Genetic algorithms as an approach to config- uration and topology design,

    C. D. Chapman, K. Saitou, and M. J. Jakiela, “Genetic algorithms as an approach to config- uration and topology design,” Journal of Mechanical Design , vol. 116, pp. 1005–1012, 1994

    work page 1994

  6. [6]

    Automatic design syn- thesis and optimization of component-based systems by evolutionary algorithms,

    P. P. Angelov, Y. Zhang, J. A. Wright, V. I. Hanby, and R. A. Buswell, “Automatic design syn- thesis and optimization of component-based systems by evolutionary algorithms,” in Genetic and Evolutionary Computation Conference , pp. 1938–1950, Springer, 2003

    work page 1938

  7. [7]

    On structural optimization of composite shell structures using a discrete constitutive parametrization,

    E. Lund and J. Stegmann, “On structural optimization of composite shell structures using a discrete constitutive parametrization,” Wind Energy, vol. 8, no. 1, pp. 109–124, 2005

    work page 2005

  8. [8]

    Configuration design of mechanical assemblies using an estimation of distribution algorithm and constraint programming,

    H. Cheong, M. Ebrahimi, A. Butscher, and F. Iorio, “Configuration design of mechanical assemblies using an estimation of distribution algorithm and constraint programming,” in IEEE Congress on Evolutionary Computation , pp. 2339–2346, 2019

    work page 2019

Show all 50 references
  1. [9]

    Multi-speed gearbox synthesis using global search and non-convex optimization,

    C. Piacentini, H. Cheong, M. Ebrahimi, and A. Butscher, “Multi-speed gearbox synthesis using global search and non-convex optimization,” in International Conference on Integration of Constraint Programming, Artificial Intelligence, and Operations Research , pp. 381–398, Springer, 2020

    2020

  2. [10]

    Truss optimization with discrete design variables: a critical review,

    M. Stolpe, “Truss optimization with discrete design variables: a critical review,” Structural and Multidisciplinary Optimization , vol. 53, no. 2, pp. 349–374, 2016

    2016

  3. [11]

    Design of planar articulated mechanisms using branch and bound,

    M. Stolpe and A. Kawamoto, “Design of planar articulated mechanisms using branch and bound,” Mathematical Programming, vol. 103, no. 2, pp. 357–397, 2005

    2005

  4. [12]

    Plane frame optimum design environment based on genetic algorithm,

    W. Jenkins, “Plane frame optimum design environment based on genetic algorithm,” Journal of Structural Engineering , vol. 118, no. 11, pp. 3103–3112, 1992

    1992

  5. [13]

    Size/geometry optimization of trusses by the force method and genetic algorithm,

    A. Kaveh and V. Kalatjari, “Size/geometry optimization of trusses by the force method and genetic algorithm,” ZAMM-Journal of Applied Mathematics and Mechanics/Zeitschrift f¨ ur Angewandte Mathematik und Mechanik: Applied Mathematics and Mechanics , vol. 84, no. 5, pp. 347–357...

    2004

  6. [14]

    Optimization of steel structures using distributed simulated annealing algorithm on a cluster of personal computers,

    H. S. Park and C. W. Sung, “Optimization of steel structures using distributed simulated annealing algorithm on a cluster of personal computers,” Computers & Structures , vol. 80, no. 14-15, pp. 1305–1316, 2002

    2002

  7. [15]

    Discrete optimization of trusses by simulated annealing,

    M. Kripka, “Discrete optimization of trusses by simulated annealing,” Journal of the Brazilian Society of Mechanical Sciences and Engineering , vol. 26, no. 2, pp. 170–173, 2004

    2004

  8. [16]

    A particle swarm ant colony optimization for truss structures with discrete variables,

    A. Kaveh and S. Talatahari, “A particle swarm ant colony optimization for truss structures with discrete variables,” Journal of Constructional Steel Research , vol. 65, no. 8-9, pp. 1558– 1568, 2009

    2009

  9. [17]

    A heuristic particle swarm optimization method for truss struc- tures with discrete variables,

    L. Li, Z. Huang, and F. Liu, “A heuristic particle swarm optimization method for truss struc- tures with discrete variables,” Computers & Structures , vol. 87, no. 7-8, pp. 435–443, 2009

    2009

  10. [18]

    The complexity of procedures for determining min- imum weight trusses with discrete member sizes,

    D. Yates, A. Templeman, and T. Boffey, “The complexity of procedures for determining min- imum weight trusses with discrete member sizes,” International Journal of Solids and Struc- tures, vol. 18, no. 6, pp. 487–495, 1982

    1982

  11. [19]

    A brief review of nature-inspired algorithms for optimization,

    I. Fister Jr, X.-S. Yang, I. Fister, J. Brest, and D. Fister, “A brief review of nature-inspired algorithms for optimization,” arXiv preprint arXiv:1307.4186 , 2013

    2013 arXiv

  12. [20]

    A novel approach to discrete truss design problems using mixed integer neighborhood search,

    M. Shahabsafa, A. Mohammad-Nezhad, T. Terlaky, L. Zuluaga, S. He, J. T. Hwang, and J. R. Martins, “A novel approach to discrete truss design problems using mixed integer neighborhood search,” Structural and Multidisciplinary Optimization , vol. 58, no. 6, pp. 2411–2429, 2018

    2018

  13. [21]

    Design optimization of dynamic flexi- ble multibody systems using the discrete adjoint variable method,

    M. Ebrahimi, A. Butscher, H. Cheong, and F. Iorio, “Design optimization of dynamic flexi- ble multibody systems using the discrete adjoint variable method,” Computers & Structures , vol. 213, pp. 82–99, 2019

    2019

  14. [22]

    Concurrent multi-scale design optimization of com- posite frame structures using the Heaviside penalization of discrete material model,

    J. Yan, Z. Duan, E. Lund, and G. Zhao, “Concurrent multi-scale design optimization of com- posite frame structures using the Heaviside penalization of discrete material model,” Acta Mechanica Sinica, vol. 32, pp. 430–441, 2016

    2016

  15. [23]

    Gradient-based selection of cross sec- tions: a novel approach for optimal frame structure design,

    C. Krogh, M. H. Jungersen, E. Lund, and E. Lindgaard, “Gradient-based selection of cross sec- tions: a novel approach for optimal frame structure design,” Structural and Multidisciplinary Optimization, vol. 56, no. 5, pp. 959–972, 2017

    2017

  16. [24]

    Discrete material selection and structural topology optimization of composite frames for maximum fundamental frequency with manu- facturing constraints,

    Z. Duan, J. Yan, I. Lee, E. Lund, and J. Wang, “Discrete material selection and structural topology optimization of composite frames for maximum fundamental frequency with manu- facturing constraints,” Structural and Multidisciplinary Optimization , vol. 60, pp. 1741–1758, 2019

    2019

  17. [25]

    A bi-level methodology for solving large-scale mixed categorical structural optimization,

    P.-J. Barjhoux, Y. Diouane, S. Grihon, D. Bettebghor, and J. Morlier, “A bi-level methodology for solving large-scale mixed categorical structural optimization,” Structural and Multidisci- plinary Optimization , vol. 62, no. 1, pp. 337–351, 2020. 30

    2020

  18. [26]

    M. P. Bendsoe and O. Sigmund, Topology optimization: theory, methods, and applications . Springer Science & Business Media, 2003

    2003

  19. [27]

    Categorical reparameterization with Gumbel-Softmax,

    E. Jang, S. Gu, and B. Poole, “Categorical reparameterization with Gumbel-Softmax,” in The International Conference on Learning Representations , 2017

    2017

  20. [28]

    The concrete distribution: A continuous relaxation of discrete random variables,

    C. Maddison, A. Mnih, and Y. Teh, “The concrete distribution: A continuous relaxation of discrete random variables,” in The International Conference on Learning Representations , 2017

    2017

  21. [29]

    E. J. Gumbel, Statistical theory of extreme values and some practical applications: a series of lectures, vol. 33. US Government Printing Office, 1954

    1954

  22. [30]

    A review of the Gumbel-max trick and its extensions for discrete stochasticity in machine learning,

    I. A. Huijben, W. Kool, M. B. Paulus, and R. J. Van Sloun, “A review of the Gumbel-max trick and its extensions for discrete stochasticity in machine learning,” IEEE Transactions on Pattern Analysis and Machine Intelligence , 2022

    2022

  23. [31]

    Boltzmann exploration done right,

    N. Cesa-Bianchi, C. Gentile, G. Lugosi, and G. Neu, “Boltzmann exploration done right,” Advances in Neural Information Processing Systems , vol. 30, 2017

    2017

  24. [32]

    Neural machine translation with Gumbel-greedy decoding,

    J. Gu, D. J. Im, and V. O. Li, “Neural machine translation with Gumbel-greedy decoding,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 32, 2018

    2018

  25. [33]

    Modeling point clouds with self-attention and Gumbel subset sampling,

    J. Yang, Q. Zhang, B. Ni, L. Li, J. Liu, M. Zhou, and Q. Tian, “Modeling point clouds with self-attention and Gumbel subset sampling,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pp. 3323–3332, 2019

    2019

  26. [34]

    Stochastic beams and where to find them: The Gumbel-top-k trick for sampling sequences without replacement,

    W. Kool, H. Van Hoof, and M. Welling, “Stochastic beams and where to find them: The Gumbel-top-k trick for sampling sequences without replacement,” in International Conference on Machine Learning , pp. 3499–3508, PMLR, 2019

    2019

  27. [35]

    V. K. Rohatgi and A. M. E. Saleh, An introduction to probability and statistics . John Wiley & Sons, 2015

    2015

  28. [36]

    A* sampling,

    C. J. Maddison, D. Tarlow, and T. Minka, “A* sampling,” Advances in Neural Information Processing Systems, vol. 27, 2014

    2014

  29. [37]

    VQ-W A V2VEC: Self-supervised learning of discrete speech representations,

    A. Baevski, S. Schneider, and M. Auli, “VQ-W A V2VEC: Self-supervised learning of discrete speech representations,” in International Conference on Learning Representations , 2019

    2019

  30. [38]

    Learning to branch for multi-task learning,

    P. Guo, C.-Y. Lee, and D. Ulbricht, “Learning to branch for multi-task learning,” in Interna- tional Conference on Machine Learning , pp. 3854–3863, 2020

    2020

  31. [39]

    Operation-aware soft channel pruning using differentiable masks,

    M. Kang and B. Han, “Operation-aware soft channel pruning using differentiable masks,” in International Conference on Machine Learning , pp. 5122–5131, 2020. 31

    2020

  32. [40]

    Adjoint sensitivity analysis for differential-algebraic equations: The adjoint dae system and its numerical solution,

    Y. Cao, S. Li, L. Petzold, and R. Serban, “Adjoint sensitivity analysis for differential-algebraic equations: The adjoint dae system and its numerical solution,” SIAM Journal on Scientific Computing, vol. 24, no. 3, pp. 1076–1089, 2003

    2003

  33. [41]

    Stacking sequence and shape optimization of laminated composite plates via a level-set method,

    G. Allaire and G. Delgado, “Stacking sequence and shape optimization of laminated composite plates via a level-set method,” Journal of the Mechanics and Physics of Solids , vol. 97, pp. 168– 196, 2016

    2016

  34. [42]

    Multimodal truss structure design using bilevel and niching based evolutionary algorithms,

    M. J. Islam, X. Li, and K. Deb, “Multimodal truss structure design using bilevel and niching based evolutionary algorithms,” in Proceedings of the Genetic and Evolutionary Computation Conference, pp. 274–281, 2017

    2017

  35. [43]

    Optimal design of continuum robots with reach- ability constraints,

    H. Cheong, M. Ebrahimi, and T. Duggan, “Optimal design of continuum robots with reach- ability constraints,” IEEE Robotics and Automation Letters , vol. 6, no. 2, pp. 3902–3909, 2021

    2021

  36. [44]

    Mine blast algorithm for opti- mization of truss structures with discrete variables,

    A. Sadollah, A. Bahreininejad, H. Eskandar, and M. Hamdi, “Mine blast algorithm for opti- mization of truss structures with discrete variables,” Computers & Structures, vol. 102, pp. 49– 63, 2012

    2012

  37. [45]

    Water cycle, mine blast and improved mine blast algorithms for discrete sizing optimization of truss structures,

    A. Sadollah, H. Eskandar, A. Bahreininejad, and J. H. Kim, “Water cycle, mine blast and improved mine blast algorithms for discrete sizing optimization of truss structures,”Computers & Structures, vol. 149, pp. 1–16, 2015

    2015

  38. [46]

    A comparative study of CBO and ECBO for optimal design of skeletal structures,

    A. Kaveh and M. I. Ghazaan, “A comparative study of CBO and ECBO for optimal design of skeletal structures,” Computers & Structures , vol. 153, pp. 137–147, 2015

    2015

  39. [47]

    An adaptive elitist differential evolution for optimization of truss structures with discrete design variables,

    V. Ho-Huu, T. Nguyen-Thoi, T. Vo-Duy, and T. Nguyen-Trang, “An adaptive elitist differential evolution for optimization of truss structures with discrete design variables,” Computers & Structures, vol. 165, pp. 59–75, 2016

    2016

  40. [48]

    R. T. Haftka and Z. G¨ urdal,Elements of structural optimization , vol. 11. Springer Science & Business Media, 2012

    2012

  41. [49]

    Computational complexity of discrete optimization prob- lems,

    J. Lenstra and A. Rinnooy Kan, “Computational complexity of discrete optimization prob- lems,” in Discrete Optimization I (P. Hammer, E. Johnson, and B. Korte, eds.), vol. 4 of Annals of Discrete Mathematics , pp. 121–140, Elsevier, 1979

    1979

  42. [50]

    J. R. R. A. Martins and A. Ning, Engineering design optimization . Cambridge University Press, Jan 2022. 32

    2022

Pith tools

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