Pith. sign in

REVIEW 5 major objections 6 minor 52 references

Adaptive Bayesian Data-Driven Design of Reliable Solder Joints for Micro-electronic Devices

T0 review · 5 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Adaptive Bayesian optimization reaches the same solder-joint design quality with half the finite-element simulations.

desk verdict Plausible adaptive BO heuristic with a real engineering case study, but the headline 3% and half-budget claims rest on single runs and a worst-case baseline. read the letter →

arxiv 2507.19663 v1 pith:5B53WJAH submitted 2025-07-25 stat.ML cs.LGphysics.comp-ph

Leo Guo , Adwait Inamdar , Willem D. van Driel , GuoQi Zhang This is my paper
classification stat.MLcs.LGphysics.comp-ph
keywords BayesianoptimizationGaussianprocessregressionadaptivehyperparametersacquisitionfunctionselectionsolderjointreliabilityaccumulatedcreepstrainSobolsensitivityanalysisfiniteelementsimulation
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

Bayesian optimization is a standard strategy for optimizing expensive simulations: build a statistical surrogate of the objective, pick the next simulation point with a so-called acquisition function, and repeat. The standard recipe fixes the surrogate's kernel and the acquisition function at the start, and this paper argues that both should instead be re-selected as data accumulate, with several acquisition-function candidates voted on at each step. On two synthetic benchmarks and a solder-joint reliability problem, the resulting adaptive scheme reaches essentially the same optimized accumulated creep strain as the best fixed-configuration Bayesian optimizer while using about half of the expensive finite-element evaluations. Because each solder-joint finite-element simulation takes 1.5 to 2 hours, halving the budget turns one optimization campaign into two and makes the reported average 3% improvement at any given budget directly valuable.

What carries the argument

The machinery is the adaptive BO loop BO-GPi-(i)Ada, which combines three mechanisms: (1) GPi, a curtailed grid search over covariance kernels and restricted likelihood domains that scores each candidate Gaussian process with the relative mean squared error on held-out data (and test log-likelihood as secondary score) and re-initializes the model periodically; (2) a candidate-design matrix built by optimizing several acquisition functions (logarithmic expected improvement, logarithmic probability of improvement, and upper confidence bound) against the same Gaussian process; and (3) iAda candidate selection, which rejects proposals that are too close to previously sampled points using an exploitation score built on the median minimum distance of the design set, then selects among the rest with a categorical probability vector that rewards acquisition functions that produced the incumbent best objective. The comparison metric WCRI (worst-case relative improvement) is what turns run-to-run noise into the paper's headline numbers.

What would settle it

Run the full five-dimensional optimization with chip rotation and CTE1 free over their whole ranges using the same 100-evaluation budget, and compare the best accumulated creep strain with the reduced 3-D results in Tables 6 and 7; if the full-space optimum beats the reduced optimum by more than about 3%, the claimed advantage is an artifact of the dimension reduction.

Watch

Extended reading notes

Core claim

The central claim is that adaptive hyperparameter handling makes Bayesian optimization more efficient than its fixed-configuration counterpart when the objective is an expensive black-box simulation. The adaptive algorithm has two components: Gaussian-process initialization (GPi), which cycles through RBF, Matérn, and rational-quadratic kernels and restricted likelihood domains, scoring each candidate model on held-out test points by relative mean squared error with test log-likelihood as a tie-breaker; and input-adaptive candidate selection, which optimizes LogEI, LogPI, and UCB proposals at each iteration, rejects candidates that fall within too small an exploitation score based on the median minimum distance to previously evaluated points, and picks among the survivors with a categorical distribution that rewards whichever acquisition found the current best objective value. On Sphere-6D and AlpineN2-3D, adaptive BO improves on the worst fixed BO configuration at most quartiles of the run; on the solder joint case, the fully adaptive scheme reaches the same accumulated creep strain as the best fixed BO with at least 50 fewer of the 100 allotted expensive objective evaluations, corresponding to an average 3% improvement at any given computational budget.

Load-bearing premise

The optimization runs only on a reduced three-dimensional design space that the authors fixed by setting the chip rotation to zero degrees and the molding compound's first thermal expansion coefficient to one of two values; if those fixed choices interact with the remaining design variables at the true optimum, the reported savings could come from having shrunk the problem rather than from the adaptive scheme.

Editorial extensions

If this is right

  • A 100-evaluation solder-joint optimization campaign can be cut to roughly 50 finite-element simulations without degrading the best design found, saving 75 to 100 hours of compute per campaign.
  • Engineers no longer need to commit in advance to one kernel and one acquisition function; the adaptive run beats the worst fixed configuration at essentially every stage of the synthetic benchmarks.
  • The per-iteration overhead of adaptive BO is about 10 to 15 seconds, which is negligible compared with the 1.5 to 2 hours needed for one finite-element simulation, so the savings are real in wall-clock terms.
  • On the solder joint case, the adaptive scheme finds designs with 2.9% lower accumulated creep strain than the best fixed BO when CTE1 is 6 ppm/°C, and an average 3.1% improvement across iterations when CTE1 is 8.5 ppm/°C.

Reading between the lines

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

  • A natural next test is to run the same adaptive loop on the full five-dimensional problem with chip rotation and CTE1 left free; if the savings persist without the dimension reduction, the mechanism is robust rather than an artifact of a smaller search space.
  • The exploitation-score and median-minimum-distance machinery could serve as a cheap stopping rule: when even the least exploitative acquisition candidate fails the threshold, further evaluations are unlikely to add information, letting a campaign end early.
  • The categorical selection rule is a minimal bandit-style reward; pairing it with explicit exploration bonuses or observed improvement magnitudes could sharpen the 3% margin on multimodal objectives.
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

5 major / 6 minor

Summary. This paper proposes BO-GPi-iAda, a Bayesian optimization heuristic that combines Gaussian-process model selection/initialization (GPi) over three kernels with an adaptive acquisition-function candidate-selection rule (SelCat/SelU) and an exploitation-score filter. The method is benchmarked against fixed-kernel/acquisition BO on Sphere-6D and AlpineN2-3D using a worst-case relative improvement measure (WCRI), and is then applied to minimize accumulated non-linear creep strain in a solder joint FEM model. The authors claim a 3% average improvement over regular BO at any budget, and that the adaptive scheme reaches the same creep-strain value as the best fixed BO with at least 50 fewer of the 100 expensive evaluations.

Significance. The practical motivation is strong: FEM evaluations cost 1.5–2 hours each, so an optimizer that reliably halves the number of evaluations would be valuable. The paper's strengths are its open-source data/code, the concrete FEM case study, the use of sensitivity analysis to reduce the design space, and a clear algorithmic description. If the comparative claims were properly supported, the contribution would be a useful heuristic addition to the BO toolbox. The current evidence, however, does not yet separate the claimed advantage from stochastic variation or from a favorable choice of baseline; the main technical claims need additional replication and sharper reporting before the result can be accepted.

major comments (5)
  1. [Section 4, Figure 5, and Eq. (13)-(14)] The headline figures (2.9%, 0.7%, 3.1%, and 'at least 50 fewer evaluations') are read from single histories of a stochastic procedure—SelCat draws the acquisition function from a categorical distribution at every iteration—yet no repeated seeds, confidence intervals, or significance tests are reported, and Table 2 reports WCRI point estimates without variance. The paper's own Future Work section (Section 5) concedes that only two synthetic functions were used. Without replication, a 3% difference between stochastic optimizers is indistinguishable from seed-to-seed variability, and the half-budget claim is not established.
  2. [Abstract and Section 2.4 vs. Eq. (20)] The comparison protocol shifts across the paper. The Abstract claims improvement over 'regular BO' at any budget, Eq. (20) defines WCRI relative to the worst quartile of the fixed-BO collection, Figure 5 uses only two Matérn-kernel baselines, and Section 4 reports 2.9% against the best vanilla BO after 100 evaluations while the '50 fewer' statement at the end of Section 4 refers to the worst-case BO scheme. Moreover, Table 2 contains negative WCRI values (e.g., AlpineN2-3D, iAda/SelCat, Q2: -18.8), which contradicts the statement that 'an adaptive BO methodology will at least perform better than the worst-case standard Bayesian optimizer.' Specify one baseline and report mean/interval results against it.
  3. [Appendix A, Figures A.8-A.12] The 5D-to-3D reduction fixes rotation at 0° using Saltelli projections obtained with CTE1=6 and CTE2=30, and fixes CTE1 at 6 or 8.5 ppm/°C based on Sobol indices. The paper never checks whether the true 5D optimum lies in the reduced domain, and it concedes that the global minimum is unknown. If rotation or CTE1 interacts with the remaining parameters away from the sampled slices, the reported improvement is an artifact of the reduced domain. Validate the reduction on the full 5D problem (e.g., a few full 5D BO runs or a validation grid) or restrict all claims to the reduced design space.
  4. [Algorithms 2 and 3 and Section 2.3] The method's own hyperparameters are underspecified. The exploitation threshold t(i) is a required input to Algorithm 3 but no values or decreasing schedule are given, and no defaults are stated for the trial threshold Q or RelMSE threshold R. Without these specifications, the reported histories cannot be reproduced from the manuscript alone. Please provide the exact schedules/defaults used and report sensitivity to them.
  5. [Algorithm 3, lines 9-10] After the ES filter, the candidate set X(i)_cand can be empty ('Skip if no candidate satisfies ES threshold'), but the algorithm then calls Sel on this set and evaluates f(x(i)), which is undefined for an empty set. Specify a fallback (e.g., accept the candidate with smallest ES, or skip the objective evaluation for that iteration and continue), since this affects the correctness of the algorithm as stated.
minor comments (6)
  1. [Abstract] The phrase 'compared any worst-performing regular Bayesian schemes' is grammatically incomplete; revise.
  2. [Algorithm 2] The RLD enumeration lists T14 = TMat(¯cmid) twice, so the subsequent indices are off by one; correct the enumeration.
  3. [Section 2.3, Eq. (13)] The sentence explaining SelCat says it rewards an acquisition for 'locating the incumbent optimum,' but the update rule rewards the selected acquisition only when the most recent selected point becomes the new incumbent; clarify the local nature of the reward.
  4. [Table 3] The text says 'the averages were recorded' while the caption reports 'median run times'; make the terminology consistent.
  5. [Figure 5] Define exactly what is plotted on the horizontal axis (number of objective evaluations, apparently) and label the vertical axis; the current caption says 'expended cost' without units.
  6. [Section 2.4] The description of BO-GPi-Ada as 'removing the exploitation score threshold step 11' appears to refer to the ES filter, which is line 9 of Algorithm 3; correct the cross-reference.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: adaptive BO is benchmarked on external synthetic functions and FEM simulation, and the reported improvements are empirical histories, not derived from fitted parameters.

full rationale

The paper's central claims—adaptive BO outperforms regular BO on synthetic objectives and reduces FEM evaluations for solder joint creep strain—are empirical benchmark results, not consequences of the method's definitions. The adaptive components (GPi kernel/hyperparameter search on held-out test points, SelCat categorical acquisition selection, and ES distance filtering) are specified independently of the performance metrics (WCRI, incumbent creep strain, number of evaluations) used to evaluate them. The 3% figure and the '50 fewer evaluations' claim are read directly from optimization histories (Fig. 5) and could have turned out differently; the paper even states that the true global minimum is unknown (Sec. 4). The dimension reduction in Appendix A uses Saltelli samples and Sobol indices as a preprocessing step that applies equally to all optimizers, and is not a prediction derived from the adaptive method. Self-references such as [17] and [18] are background or a passing remark, not load-bearing evidence for the improvement claim. No equation in the paper reduces to its own input, and no fitted parameter is renamed as a prediction.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The paper introduces a heuristic method with several free thresholds and design choices. No new physical entities are introduced. The central engineering claim depends on the arbitrary reduction of the design space and on thresholds that are not quantified in the manuscript.

free parameters (6)
  • Exploitation score threshold t(i) = unspecified
    Controls when candidate designs are rejected as too close to previous samples. The text only says it is programmed to decrease with i, with no formula or chosen values.
  • RelMSE threshold R = 0.05
    Stopping criterion in Algorithm 2 for GP model selection. Chosen without sensitivity analysis.
  • Trial threshold Q = unspecified
    Caps the number of GP model candidates in the GPi search. No value or sensitivity study is provided.
  • Nominal hyperparameter fixture values = low, mid, high (values not given)
    Used to build restricted likelihood domains for covariance kernel parameters. The actual numerical values are not stated.
  • Train-test split 20:80 = Ntest/N = 1/5
    Fixed as the split used throughout the paper. No analysis shows that this split is optimal or robust.
  • CTE1 nominal values 6 and 8.5 ppm/degree C = 6 ppm/degree C and 8.5 ppm/degree C
    Selected based on perceived variance in Saltelli sample projections in Appendix A. This choice reduces the design space from 5D to 3D and is load-bearing for the case study results.
assumptions (4)
  • domain assumption The expensive objective evaluations are noiseless, so RelMSE is an exact quality measure for the GP surrogate.
    Invoked in Section 2.2, Eq. (8), where the paper states 'If f is assumed noiseless, i.e. f(Xtest)=ytest exactly...'
  • domain assumption Evaluating the objective f is orders of magnitude more expensive than the inner optimization loops, justifying extra per-iteration computation.
    Stated in Section 2.1 as the assumption underpinning the motivation to use BO at all.
  • ad hoc to paper The kernel set {RBF, Matérn, RQ} and acquisition set {LogEI, LogPI, UCB} are sufficient for good BO performance.
    The paper restricts to K=3 kernels and A=3 acquisition functions without evidence that this choice covers the useful space.
  • ad hoc to paper Fixing rotation to 0 degrees and CTE1 to 6 or 8.5 ppm/degree C preserves the optimum of the accumulated creep strain.
    Assumed in Appendix A based on Saltelli sample projections and Sobol indices, but the full 5D optimum is never verified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Adaptive Bayesian Data-Driven Design of Reliable Solder Joints for Micro-electronic Devices." pith.science (2026). https://pith.science/paper/5B53WJAH

@misc{pith2026250719663,
  author       = {Pith},
  title        = {Pith review of: Adaptive Bayesian Data-Driven Design of Reliable Solder Joints for Micro-electronic Devices},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5B53WJAH}},
  note         = {Machine review of arXiv:2507.19663}
}
read the original abstract

Solder joint reliability related to failures due to thermomechanical loading is a critically important yet physically complex engineering problem. As a result, simulated behavior is oftentimes computationally expensive. In an increasingly data-driven world, the usage of efficient data-driven design schemes is a popular choice. Among them, Bayesian optimization (BO) with Gaussian process regression is one of the most important representatives. The authors argue that computational savings can be obtained from exploiting thorough surrogate modeling and selecting a design candidate based on multiple acquisition functions. This is feasible due to the relatively low computational cost, compared to the expensive simulation objective. This paper addresses the shortcomings in the adjacent literature by providing and implementing a novel heuristic framework to perform BO with adaptive hyperparameters across the various optimization iterations. Adaptive BO is subsequently compared to regular BO when faced with synthetic objective minimization problems. The results show the efficiency of adaptive BO when compared any worst-performing regular Bayesian schemes. As an engineering use case, the solder joint reliability problem is tackled by minimizing the accumulated non-linear creep strain under a cyclic thermal load. Results show that adaptive BO outperforms regular BO by 3% on average at any given computational budget threshold, critically saving half of the computational expense budget. This practical result underlines the methodological potential of the adaptive Bayesian data-driven methodology to achieve better results and cut optimization-related expenses. Lastly, in order to promote the reproducibility of the results, the data-driven implementations are made available on an open-source basis.

Figures

Figures reproduced from arXiv: 2507.19663 by the authors.

Figure 1
Figure 1. Two-dimensional design space samples (black dots) with disks (gray) of radius MMD. [PITH_FULL_IMAGE:figures/full_fig_p015_1.png] view at source ↗
Figure 2
Figure 2. Flowchart diagram of BO-GPi-(i)Ada, Algorithm [PITH_FULL_IMAGE:figures/full_fig_p017_2.png] view at source ↗
Figure 3
Figure 3. Five-number summary of BO (Algorithm 1) and BO-GPi-iAda optimization histories (Algorithm 3). A comparison between BO, with the Mat´ern covariance kernel and the logarithmic Expected Improvement acquisition function as chosen hyperparameters, and BO-GPi-iAda with two different candidate selection methods. The incumbent minimal objective value is plotted against the iteration. The vertical gray dashed lines indicate … view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Schematic overview of the PCB model, the electronic package submodel and the solder [PITH_FULL_IMAGE:figures/full_fig_p021_4.png]
Figure 5
Figure 5. Figure 5: Accumulated non-linear creep strain optimization histories, with the incumbent min [PITH_FULL_IMAGE:figures/full_fig_p023_5.png]
Figure 6
Figure 6. Figure 6: Accumulated creep profiles across the critical solder joint for selected designs (CTE1 [PITH_FULL_IMAGE:figures/full_fig_p024_6.png]
Figure 7
Figure 7. Figure 7: Accumulated creep profiles across the critical solder joint for selected designs (CTE1 [PITH_FULL_IMAGE:figures/full_fig_p025_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

52 extracted references · 46 canonical work pages

  1. [1]

    K. Guo, Z. Yang, C.-H. Yu, M. J. Buehler, Artificial intelligence and machine learning in design of mechanical materials, Materials Horizons (2021). 31

    work page 2021

  2. [2]

    Shahriari, K

    B. Shahriari, K. Swersky, Z. Wang, R. P. Adams, N. De Freitas, Taking the human out of the loop: A review of bayesian optimization, Proceedings of the IEEE 104 (1) (2015) 148–175

    work page 2015

  3. [3]

    Garnett, Bayesian optimization, Cambridge University Press, 2023

    R. Garnett, Bayesian optimization, Cambridge University Press, 2023

    2023

  4. [4]

    Zhang, W

    S. Zhang, W. Lyu, F. Yang, C. Yan, D. Zhou, X. Zeng, X. Hu, An efficient multi-fidelity bayesian optimization approach for analog circuit synthesis, in: 2019 56th ACM/IEEE Design Automation Conference (DAC), IEEE, 2019, pp. 1–6

    work page 2019

  5. [5]

    Huang, C

    J. Huang, C. Tao, F. Yang, C. Yan, D. Zhou, X. Zeng, Bayesian optimiza- tion approach for rf circuit synthesis via multitask neural network enhanced gaussian process, IEEE Transactions on Microwave Theory and Techniques 70 (11) (2022) 4787–4795

    work page 2022

  6. [6]

    Zhang, F

    S. Zhang, F. Yang, C. Yan, D. Zhou, X. Zeng, An efficient batch-constrained bayesian optimization approach for analog circuit synthesis via multiobjec- tive acquisition ensemble, IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems 41 (1) (2021) 1–14

    work page 2021

  7. [7]

    Tilgner, Physics of failure for interconnect structures: an essay, Microsys- tem technologies 15 (1) (2009) 129–138

    R. Tilgner, Physics of failure for interconnect structures: an essay, Microsys- tem technologies 15 (1) (2009) 129–138

    work page 2009

  8. [8]

    X. Li, R. Sun, Y. Wang, A review of typical thermal fatigue failure models for solder joints of electronic components, in: IOP Conference Series: Materials Science and Engineering, Vol. 242, IOP Publishing, 2017, p. 012103

    work page 2017

Show all 52 references
  1. [9]

    Musadiq, W

    M. Musadiq, W. D. van Driel, R. Roucou, R. Rongen, G. Zhang, Impact of ip block placement on solder joint reliability in ic packages, in: 2025 26th Inter- national Conference on Thermal, Mechanical and Multi-Physics Simulation and Experiments in Microelectronics and Microsyste...

    2025

  2. [10]

    Pecht, Handbook of electronic package design, CRC press, 2018

    M. Pecht, Handbook of electronic package design, CRC press, 2018

    2018

  3. [11]

    Wolfgang, Examples for failures in power electronics systems, ECPE tu- torial on reliability of power electronic systems, Nuremberg, Germany (2007) 19–20

    E. Wolfgang, Examples for failures in power electronics systems, ECPE tu- torial on reliability of power electronic systems, Nuremberg, Germany (2007) 19–20

    2007

  4. [12]

    O. O. Ogbomo, E. H. Amalu, N. Ekere, P. Olagbegi, Effect of operating temperature on degradation of solder joints in crystalline silicon photovoltaic modules for improved reliability in hot climates, Solar Energy 170 (2018) 682–693. 32

    2018

  5. [13]

    Arabi, A

    F. Arabi, A. Gracia, J.-Y. Del´ etage, H. Fr´ emont, Effect of thermal and vibra- tional combined ageing on qfn terminal pads solder reliability, Microelectron- ics Reliability 114 (2020) 113883

    2020

  6. [14]

    E.-H. Wong, W. Van Driel, A. Dasgupta, M. Pecht, Creep fatigue models of solder joints: A critical review, Microelectronics Reliability 59 (2016) 1–12

    2016

  7. [15]

    De Jong, A

    S. De Jong, A. Ghezeljehmeidan, W. Van Driel, Physics-informed machine learning for solder joint qualification tests, in: 2024 25th International Confer- ence on Thermal, Mechanical and Multi-Physics Simulation and Experiments in Microelectronics and Microsystems (EuroSimE), I...

    2024

  8. [16]

    de Jong, A

    S. de Jong, A. Ghezeljehmeidan, W. van Driel, Solder joint reliability predic- tions using physics-informed machine learning, Microelectronics Reliability 172 (2025) 115797

    2025

  9. [17]

    Van Driel, G

    W. Van Driel, G. Zhang, J. Janssen, L. Ernst, Response surface modeling for nonlinear packaging stresses, J. Electron. Packag. 125 (4) (2003) 490–497

    2003

  10. [18]

    Wymys lowski, W

    A. Wymys lowski, W. Van Driel, J. Van De Peer, N. Tzannetakis, G. Zhang, Advanced numerical prototyping methods in modern engineering applications–optimisation for micro-electronic package reliability, Microelec- tronics Reliability 47 (2-3) (2007) 280–289

    2007

  11. [19]

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

    2018 arXiv

  12. [20]

    Feurer, J

    M. Feurer, J. T. Springenberg, F. Hutter, Initializing bayesian hyperparam- eter optimization via meta-learning, in: Twenty-Ninth AAAI Conference on Artificial Intelligence, 2015

    2015

  13. [21]

    L. Yang, A. Shami, On hyperparameter optimization of machine learning algorithms: Theory and practice, Neurocomputing 415 (2020) 295–316

    2020

  14. [22]

    D. Ueda, A. Yamamoto, T. Takashima, N. Onoda, S. Noda, S. Kashiwagi, T. Morisaki, T. Honjo, A. Shimazaki, Y. Miki, Training, validation, and test of deep learning models for classification of receptor expressions in breast cancers from mammograms, JCO Precision Oncology 5 (202...

    2021

  15. [23]

    Khalil, A

    M. Khalil, A. S. McGough, Z. Pourmirza, M. Pazhoohesh, S. Walker, Machine learning, deep learning and statistical analysis for forecasting building energy consumption—a systematic review, Engineering Applications of Artificial In- telligence 115 (2022) 105287. 33

    2022

  16. [24]

    Sharma, T

    A. Sharma, T. Mukhopadhyay, S. M. Rangappa, S. Siengchin, V. Kushvaha, Advances in computational intelligence of polymer composite materials: ma- chine learning assisted modeling, analysis and design, Archives of Computa- tional Methods in Engineering 29 (5) (2022) 3341–3385

    2022

  17. [25]

    C. K. Williams, C. E. Rasmussen, Gaussian processes for machine learning, Vol. 2, MIT press Cambridge, MA, 2006

    2006

  18. [26]

    D. R. Jones, M. Schonlau, W. J. Welch, Efficient global optimization of ex- pensive black-box functions, Journal of Global optimization 13 (4) (1998) 455–492

    1998

  19. [27]

    H. J. Kushner, A new method of locating the maximum point of an arbitrary multipeak curve in the presence of noise (1964)

    1964

  20. [28]

    Auer, Using confidence bounds for exploitation-exploration trade-offs, Jour- nal of Machine Learning Research 3 (Nov) (2002) 397–422

    P. Auer, Using confidence bounds for exploitation-exploration trade-offs, Jour- nal of Machine Learning Research 3 (Nov) (2002) 397–422

    2002

  21. [29]

    D. P. Kingma, J. Ba, Adam: A method for stochastic optimization, arXiv preprint arXiv:1412.6980 (2014)

    2014 arXiv

  22. [30]

    C. Zhu, R. H. Byrd, P. Lu, J. Nocedal, Algorithm 778: L-bfgs-b: Fortran subroutines for large-scale bound-constrained optimization, ACM Transac- tions on Mathematical Software (TOMS) 23 (4) (1997) 550–560

    1997

  23. [31]

    Ament, S

    S. Ament, S. Daulton, D. Eriksson, M. Balandat, E. Bakshy, Unexpected improvements to expected improvement for bayesian optimization, arXiv preprint arXiv:2310.20708 (2023)

    2023 arXiv

  24. [32]

    Duvenaud, J

    D. Duvenaud, J. Lloyd, R. Grosse, J. Tenenbaum, G. Zoubin, Structure dis- covery in nonparametric regression through compositional kernel search, in: International Conference on Machine Learning, 2013, pp. 1166–1174

    2013

  25. [33]

    Chicco, M

    D. Chicco, M. J. Warrens, G. Jurman, The coefficient of determination r- squared is more informative than smape, mae, mape, mse and rmse in regres- sion analysis evaluation, Peerj computer science 7 (2021) e623

    2021

  26. [34]

    S. K. Deshpande, S. Ghosh, T. D. Nguyen, T. Broderick, Are you using test log-likelihood correctly?, arXiv preprint arXiv:2212.00219 (2022)

    2022 arXiv

  27. [35]

    Gelman, J

    A. Gelman, J. Hwang, A. Vehtari, Understanding predictive information cri- teria for bayesian models, Statistics and computing 24 (2014) 997–1016. 34

    2014

  28. [36]

    Watanabe, Tree-structured parzen estimator: Understanding its algorithm components and their roles for better empirical performance, arXiv preprint arXiv:2304.11127 (2023)

    S. Watanabe, Tree-structured parzen estimator: Understanding its algorithm components and their roles for better empirical performance, arXiv preprint arXiv:2304.11127 (2023)

    2023 arXiv

  29. [37]

    Hutter, H

    F. Hutter, H. H. Hoos, K. Leyton-Brown, Sequential model-based optimiza- tion for general algorithm configuration, in: International conference on learn- ing and intelligent optimization, Springer, 2011, pp. 507–523

    2011

  30. [38]

    Bergstra, B

    J. Bergstra, B. Komer, C. Eliasmith, D. Yamins, D. D. Cox, Hyperopt: a python library for model selection and hyperparameter optimization, Com- putational Science & Discovery 8 (1) (2015) 014008

    2015

  31. [39]

    Akiba, S

    T. Akiba, S. Sano, T. Yanase, T. Ohta, M. Koyama, Optuna: A next- generation hyperparameter optimization framework, in: Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2019

    2019

  32. [40]

    A. G. Barto, Reinforcement learning, in: Neural systems for control, Elsevier, 1997, pp. 7–30

    1997

  33. [41]

    Kandasamy, K

    K. Kandasamy, K. R. Vysyaraju, W. Neiswanger, B. Paria, C. R. Collins, J. Schneider, B. Poczos, E. P. Xing, Tuning hyperparameters without grad students: Scalable and robust bayesian optimisation with dragonfly (2019). arXiv:1903.06694

    2019 arXiv

  34. [42]

    I. M. Sobol, Global sensitivity indices for nonlinear mathematical models and their monte carlo estimates, Mathematics and computers in simulation 55 (1-

  35. [43]

    W. K. Hastings, Monte carlo sampling methods using markov chains and their applications (1970)

    1970

  36. [44]

    M. D. McKay, R. J. Beckman, W. J. Conover, A comparison of three methods for selecting values of input variables in the analysis of output from a computer code, Technometrics 42 (1) (2000) 55–61

    2000

  37. [45]

    Saltelli, P

    A. Saltelli, P. Annoni, I. Azzini, F. Campolongo, M. Ratto, S. Tarantola, Variance based sensitivity analysis of model output. design and estimator for the total sensitivity index, Computer physics communications 181 (2) (2010) 259–270

    2010

  38. [46]

    Niessner, A

    M. Niessner, A. Gyarmati, H. Guettler, Impact of mechanical material mod- eling on the solder joint fatigue analysis of a leadless package mounted at different positions inside a generic aluminum ecu, in: 2023 24th International 35 Conference on Thermal, Mechanical and Multi-P...

    2023

  39. [47]

    Gromala, A

    P. Gromala, A. Inamdar, A. Prisacaru, M. Dressler, A. Kabakchiev, Degra- dation and remaining useful life prediction of automotive electronics, in: Reliability of Organic Compounds in Microelectronics and Optoelectronics, Springer, 2022, pp. 415–447

    2022

  40. [48]

    Nossent, P

    J. Nossent, P. Elsen, W. Bauwens, Sobol’sensitivity analysis of a complex environmental model, Environmental Modelling & Software 26 (12) (2011) 1515–1525

    2011

  41. [49]

    Zhang, M

    X.-Y. Zhang, M. N. Trame, L. J. Lesko, S. Schmidt, Sobol sensitivity analysis: a tool to guide the development and evaluation of systems pharmacology models, CPT: pharmacometrics & systems pharmacology 4 (2) (2015) 69–79

    2015

  42. [50]

    D. Yang, G. Zhang, W. Van Driel, J. Janssen, H. Bressers, L. Ernst, Parameter sensitivity study of cure-dependent underfill properties on flip chip failures, in: 52nd Electronic Components and Technology Conference 2002.(Cat. No. 02CH37345), IEEE, 2002, pp. 865–872

    2002

  43. [51]

    D.-G. Yang, J. Liang, Q.-Y. Li, L. J. Ernst, G. Zhang, Parametric study on flip chip package with lead-free solder joints by using the probabilistic designing approach, Microelectronics Reliability 44 (12) (2004) 1947–1955

    2004

  44. [52]

    Vandevelde, E

    B. Vandevelde, E. Beyne, K. G. Zhang, J. F. Caers, D. Vandepitte, M. Baelmans, Solder parameter sensitivity for csp life-time prediction us- ing simulation-based optimization method, IEEE transactions on electronics packaging manufacturing 25 (4) (2003) 318–325. 36

    2003

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.