Pith. sign in

REVIEW 3 major objections 4 minor 41 references

Efficient Fireworks Algorithm Equipped with an Explosion Mechanism based on Student's T-distribution

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A fireworks algorithm that samples from a Student's t-distribution, with updates derived from natural gradient descent, claims to match or beat state-of-the-art optimizers on standard benchmarks.

desk verdict A credible empirical FWA variant with an over-sold natural-gradient story; the benchmarks deserve a referee, but the 'solid theoretical foundation' claim needs to be walked back or quantified. read the letter →

arxiv 2506.08484 v1 pith:NA4ML37X submitted 2025-06-10 cs.NE

Cen Shipeng , Tan Ying This is my paper
classification cs.NE
keywords fireworksalgorithmstudent'st-distributionnaturalgradientdescentexploration-exploitationbalancederivative-freeoptimizationCEC2013CEC2017heavy-tailedsampling
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 claims that the fireworks algorithm's explosion step—traditionally uniform sampling in a hypercube—can be replaced by sampling from a Student's t-distribution whose mean, scale matrix, and degrees of freedom are updated by natural gradient descent. The t-distribution is positioned as a tunable middle ground between Gaussian and Cauchy sampling: raising the degrees of freedom makes the sampler behave like CMA-ES (strong local exploitation), while low degrees of freedom preserve heavy-tailed exploration. With an adaptive degree-of-freedom rule and a restart mechanism, the resulting algorithm, TFWA, is claimed to be the strongest fireworks-algorithm variant and to match or beat state-of-the-art optimizers on the CEC2013 and CEC2017 benchmarks, including a qualitative breakthrough on unimodal problems. If correct, this would give a principled, parameter-light way to add heavy-tailed exploration to derivative-free optimizers and extend the information-geometric view of CMA-ES to a whole family of samplers.

What carries the argument

The load-bearing object is the Fisher information matrix of the multivariate Student's t-distribution, equation (17), and its inverse. To make the inverse tractable the paper drops the trace-trace cross term, leaving the per-block inverses $F_m^{-1}=\frac{d+v+2}{d+v}\Sigma$ and the corresponding scale update (47); these, fused with CMA-ES-style rank weights, define the explosion weights of Algorithm 1. A second mechanism, AdjustDegreeOfFreedom, increases the degrees of freedom when a better solution is found, letting the sampler interpolate between heavy-tailed exploration and Gaussian-like exploitation, while a loser-out tournament restarts unpromising fireworks.

What would settle it

Run TFWA on a rotated ill-conditioned quadratic such as the CEC elliptic function and record, at each generation, the magnitude of the dropped term $-\frac{1}{2(d+v+2)}\operatorname{tr}(\partial\Sigma/\partial\theta_i\,\Sigma^{-1})\operatorname{tr}(\partial\Sigma/\partial\theta_j\,\Sigma^{-1})$ in equation (40) relative to the kept curvature term. If it is not at least an order of magnitude smaller, the inverse-Fisher update is not the correct natural gradient, contradicting the paper's stated theoretical foundation.

Watch

Extended reading notes

Core claim

The central discovery is that a Student's t-distribution explosion operator can be derived from natural gradient descent, not just grafted on heuristically. Starting from the t-density and its Fisher information matrix, the paper obtains natural-gradient update rules for the mean and the scale matrix, equations (44) and (47), which generalize the familiar Gaussian updates of CMA-ES and reduce to them as the degrees of freedom grow. The algorithm then fuses these natural-gradient weights with rank-based weights, so each sampled spark contributes to reshaping the sampling distribution in proportion to its fitness. The adaptive degrees-of-freedom rule completes the mechanism: when a firework improves, the sampler becomes more Gaussian and exploits; when it stagnates, the heavy tails remain and exploration continues. The paper's claim is that this combination makes TFWA the strongest fireworks-algorithm variant and competitive with state-of-the-art optimizers on CEC2013 and CEC2017, with particular strength on unimodal problems and on functions with many extreme points.

Load-bearing premise

The derivation's validity rests on treating the trace-trace term of the Fisher information matrix as negligible; if the scale matrix becomes ill-conditioned, that term can be large and the resulting update is no longer the true natural gradient.

Editorial extensions

If this is right

  • TFWA reaches the global optimum on unimodal and convex problems in the CEC suites, closing the gap that previously kept fireworks-algorithm variants far behind SOTA on such problems.
  • The degree-of-freedom adaptation automatically shifts the sampler from heavy-tailed exploration to Gaussian-like exploitation, making the separate guiding-spark mutation operator unnecessary.
  • On multimodal CEC2017 functions with many local extrema, TFWA beats Gaussian-sampling algorithms such as RB-IPOP-CMA-ES, indicating that heavy-tailed explosion sampling helps escape many local optima.
  • On the CEC2017 benchmark TFWA places second to the competition winner EBOwithCMAR and ahead of two other CMA-based SOTA variants, while on CEC2013 30-D it achieves the best average rank among all compared algorithms.

Reading between the lines

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

  • Because the trace-trace approximation has no error bound, TFWA's claimed 'solid theoretical foundation' likely holds in the regime where the scale matrix is near-isotropic or the degrees of freedom are large; on strongly rotated, ill-conditioned landscapes, the empirical gains may come as much from the restart and degree-of-freedom schedules as from the natural-gradient correction. This is an edit
  • The same derivation could be applied to other heavy-tailed families or to other swarm optimizers that currently sample with Gaussians, giving each a tunable exploration tail without changing the underlying selection skeleton.
  • A direct ablation—running TFWA with the full Fisher inverse instead of the approximated one on rotated CEC functions—would settle whether the approximation is actually harmless in practice.
  • The trajectory of the degrees of freedom across a run could serve as a cheap diagnostic of landscape difficulty: functions where the degrees of freedom grow slowly are the ones where the optimizer is still exploring.
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

3 major / 4 minor

Summary. The paper proposes TFWA, a fireworks algorithm variant whose explosion sampling uses a multivariate Student's t-distribution. The distribution parameters are updated using a natural-gradient-descent-style derivation for the t-distribution, with a degree-of-freedom adaptation rule and a loser-out restart mechanism adopted from LoTFWA. The algorithm is evaluated on CEC2013 (30 and 50 dimensions) and CEC2017 (30 dimensions) with a budget of 10,000d evaluations and 30 independent runs, comparing against two FWA variants and several state-of-the-art competitors. The central claims are that TFWA is the strongest FWA variant, that it matches or surpasses top CEC2013/CEC2017 algorithms on average rank, and that the t-distribution sampling gives a theoretically founded generalization of Gaussian-based CMA-ES sampling.

Significance. If the claims hold, TFWA would be a notable contribution: it extends the information-geometric derivation of CMA-ES to a tunable heavy-tailed family, demonstrates that a t-distribution sampler can beat Gaussian-based CMA-ES variants on multimodal problems, and strengthens the FWA line of research. The experimental study is substantial: standard evaluation budget, 30 independent runs, one-to-one Wilcoxon tests, and per-function tables plus convergence plots. The derivation is presented step-by-step in the appendix, which is valuable even where it needs repair. The paper's main weakness is that the load-bearing theoretical approximation (dropping the trace-trace Fisher term) is not justified with any error bound, and the pseudo-code has notational/indexing errors that prevent exact reproduction. With those fixed, the contribution would be appropriate for the journal.

major comments (3)
  1. [Appendix, Eq. (40) and (42)-(47)] The derivation of the inverse Fisher information for the scale matrix relies on setting the trace-trace term -1/(2(d+v+2)) tr(∂Σ/∂θ_i Σ^{-1}) tr(∂Σ/∂θ_j Σ^{-1}) to zero. The argument given after Eq. (40) (bounded two-norm of the transformation matrix, d+v+2 ≥ v, and large empirical v) does not bound the size of this term relative to the kept term. On rotated ill-conditioned CEC functions, Σ can have condition number 10^6 or larger, and the trace-trace term along directions associated with small eigenvalues can be comparable to or larger than the kept term. Because Eqs. (42), (44), (45), and (47) all inherit this approximation, the claimed natural-gradient update for the t-distribution is not established. Please provide an explicit error bound, or a numerical test measuring the Frobenius norm of the dropped term relative to the retained block along a representative TFWA trajectory on, for example, CEC2013 f15/f16 or CEC2017 f10.
  2. [Algorithm 1, lines 8, 11-13] The pseudo-code as written is not executable and the update is not translation invariant. Line 8 computes w'_i using pop[i]^T C^{-1} pop[i] without centering on the current mean m; the theory in Eq. (44) uses (x-m). Line 11 m' ← w'' * pop is a weighted mean only if pop is sorted and the product is a sum over the λ samples, which is not stated; as written it is a matrix expression. Line 12 sets cov_i to the scalar (pop_i - m_i)^T (pop_i - m_i), whereas the covariance update requires the outer product (pop_i - m_i)(pop_i - m_i)^T. In addition, the rank-based weights w_i in line 2 are meaningful only after sorting the sampled population by fitness, but no fitness-sorting step appears before line 9. These issues make the algorithm impossible to reproduce exactly and must be corrected with a precise, self-contained pseudocode.
  3. [Tables VI-VIII and Section IV.B] The comparison against NBIPOPaCMA, SHADE, MVMO, SPSO2011, EBOwithCMAR, LSHADE-SPACMA, and RB-IPOP-CMA-ES is presented as evidence that TFWA 'matches or surpasses SOTA' and is 'far superior' on some functions. The paper does not state whether the baseline results were re-run under the same conditions (budget, initialization, bound handling, number of runs) or taken from published competition reports. If the latter, the comparison is not controlled: CEC competition results often use different population sizes, restarts, and evaluation budgets. Please specify the provenance of every baseline column and, ideally, provide local re-runs with identical experimental protocol, or explicitly restrict the claim to a comparison against previously reported numbers.
minor comments (4)
  1. [Section II.B, Eqs. (3)-(4)] The tail bounds are derived only for a ≥ 0; please state that the argument is symmetric and valid for |a| or specify the one-sided tail being bounded.
  2. [Appendix, paragraph after Eq. (40)] The sentence 'F is a diagonal matrix' is inaccurate; the Fisher information is block-structured, and the covariance block is not diagonal in general. Please rephrase as 'block-diagonal between the mean and scale parameters' or qualify the statement.
  3. [Algorithm 2, step 2] The update df' ← max(df*factor, df+1) is unusual because for small df the df+1 branch dominates; please clarify whether the intent is to guarantee a minimum increase of 1 and describe the rationale.
  4. [Throughout] The paper contains inconsistent spacing in 'FW A' and inconsistent capitalization of 'student's T'; please standardize these terms.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the t-distribution natural-gradient derivation is self-contained, and the performance claims are tested against external CEC benchmarks rather than being constructed from fitted values.

full rationale

The paper's central theoretical derivation computes the Fisher information of the multivariate Student's t-distribution from the density in Eq. (5) using explicit integrals in the appendix (Eqs. 20-47). The resulting natural-gradient weights (Eqs. 18-19 and 44, 47) are derived from the distributional model itself, not from benchmark optima or fitted parameters. The dropped trace-trace term in Eq. (40) is an acknowledged approximation described as 'for the convenience of the theoretical derivation'; this is a rigor and correctness concern (no error bound is provided), but it is not a circular step because the approximation does not assume the conclusion. The self-citations present in the paper, such as adopting the LoTFWA restart strategy [11] and using LoTFWA/MGFWA [11,12] as baselines, are algorithmic components and comparative references; they are not used to justify the central theoretical claim, and the benchmarks (CEC2013, CEC2017) are external and independent of the authors' fitted values. The degree-of-freedom adjustment heuristic is presented without an ablation, but it is not called a prediction nor is it derived from the benchmarks. No equation in the paper reduces to a fitted input or to a self-citation chain, so the derivation chain is not circular.

Assumptions & free parameters 5 free parameters · 7 assumptions · 0 invented entities

The central algorithm is assembled from a known distribution family, a known gradient-preconditioning principle, and two hand-designed mechanisms. The reader is asked to accept the trace-trace approximation, the heuristic fusion of natural-gradient and rank-based weights, and the df-adaptation and restart rules without error bounds or ablation; these are the true price of entry for the claimed benchmark performance.

free parameters (5)
  • number of fireworks N = 2
    Chosen in Section IV experimental setup; no sensitivity analysis provided.
  • initial degrees of freedom df = 5
    Chosen in Section IV; controls tail weight of explosion sampling; no sensitivity analysis.
  • degree-of-freedom factor array = [1.05, 10]
    Used in Algorithm 2 to grow df when fitness improves; the second value is not clearly explained; hand-selected.
  • sparks per generation lambda = 10d/N
    Evaluation budget parameter scaling with dimension; inherited from FWA practice; no analysis of trade-off.
  • CMA-ES-style learning rates and scale parameters = Table I formulas
    Adopted from CMA-ES and prior FWA; treated as constants rather than tuned per problem.
assumptions (7)
  • standard math Multivariate Student's t-distribution density and its moments (Eq. 5-7)
    Used as the sampling distribution and for the Fisher information calculation; standard result.
  • domain assumption Natural gradient descent framework of Amari and its CMA-ES interpretation (Akimoto et al.)
    Justifies the use of the inverse Fisher information to precondition gradient updates; assumes the objective can be estimated via samples from the search distribution.
  • ad hoc to paper Trace-trace term of the Fisher information for covariance is negligible
    Stated in the appendix after Eq. (40); used to invert the Fisher information and obtain Eq. (18) and (19); no error bound is given.
  • ad hoc to paper Fusing rank-based weights with natural-gradient weights produces a valid update
    Algorithm 1 lines 9 and 13; heuristic design choice, not derived.
  • ad hoc to paper Degree-of-freedom adaptation rule
    Algorithm 2; not derived from the NGD framework; the schedule is a heuristic.
  • domain assumption Loser-out tournament restart from LoTFWA
    Adopted from the authors' prior work [11], not re-derived here.
  • ad hoc to paper Removal of the guiding-spark mutation operator is beneficial
    Argued in Sections III and IV from between-algorithm comparisons; no component ablation supports it.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Efficient Fireworks Algorithm Equipped with an Explosion Mechanism based on Student's T-distribution." pith.science (2026). https://pith.science/paper/NA4ML37X

@misc{pith2026250608484,
  author       = {Pith},
  title        = {Pith review of: Efficient Fireworks Algorithm Equipped with an Explosion Mechanism based on Student's T-distribution},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NA4ML37X}},
  note         = {Machine review of arXiv:2506.08484}
}
read the original abstract

Many real-world problems can be transformed into optimization problems, which can be classified into convex and non-convex. Although convex problems are almost completely studied in theory, many related algorithms to many non-convex problems do not work well and we need more optimization techniques. As a swarm intelligence optimization algorithm, the Fireworks Algorithm(FWA) has been widely studied and applied to many real-world scenarios, even including large language model fine-tuning. But the current fireworks algorithm still has a number of problems. Firstly, as a heuristic algorithm, its performance on convex problems cannot match the SOTA results, and can even be said to be unsatisfactory; secondly, the sampling methods (explosion) of most FWA variants are still uniform sampling, which is actually inefficient in high dimensional cases. This work of ours proposes a new student's t-distribution based FWA(TFWA) with a solid theoretical foundation, which fully utilizes the advantage that student's t-distribution can adjust the parameters (degrees of freedom) and thus adjust the exploitation capability. We have fully experimented on mainstream benchmarks CEC2013 and CEC2017, which proves that TFWA not only becomes the strongest variant of the fireworks algorithm, but also achieves results comparable to SOTA on the test set, and its performance is far superior to that of the SOTA algorithm in some scenarios with a large number of extreme points.

Figures

Figures reproduced from arXiv: 2506.08484 by the authors.

Figure 1
Figure 1. Standard Gaussian and Cauchy Distribution [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Visualization of TFWA optimization progress on func [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. CEC2013 f1-f8 [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: CEC2013 f9-f16 [PITH_FULL_IMAGE:figures/full_fig_p015_4.png]
Figure 5
Figure 5. Figure 5: CEC2013 f17-f24 [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]
Figure 6
Figure 6. Figure 6: CEC2013 f25-f28 [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: CEC2017 f1-f8 [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]
Figure 8
Figure 8. Figure 8: CEC2017 f9-f16 [PITH_FULL_IMAGE:figures/full_fig_p019_8.png]
Figure 9
Figure 9. Figure 9: CEC2017 f17-f24 [PITH_FULL_IMAGE:figures/full_fig_p020_9.png]
Figure 10
Figure 10. Figure 10: CEC2017 f25-f30 [PITH_FULL_IMAGE:figures/full_fig_p021_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

41 extracted references · 39 canonical work pages

  1. [1]

    Gao, J., Li, P., Chen, Z., & Zhang, J. (2020). A survey on deep learning for multimodal data fusion. Neural Computation, 32(5), 829-864

    work page 2020

  2. [2]

    H., & Chan, C

    Nguyen, H. H., & Chan, C. W. (2006). Applications of artificial intelligence for optimization of compressor scheduling. Engineering Applications of Artificial Intelligence, 19(2), 113-126

    work page 2006

  3. [3]

    Artificial intel732 ligence applied to stock market trading: a review

    work page

  4. [4]

    H., & Hutter, F

    Elsken, T., Metzen, J. H., & Hutter, F. (2019). Neural Architecture Search: A Survey, Journal of Machine Learning Research

    work page 2019

  5. [5]

    Holland, J. H. (1992). Genetic algorithms. Scientific american, 267(1), 66-73

    work page 1992

  6. [6]

    Marini, F., & Walczak, B. (2015). Particle swarm optimization (PSO). A tutorial. Chemometrics and Intelligent Laboratory Systems, 149, 153- 165

    work page 2015

  7. [7]

    Dorigo, M., Birattari, M., & Stutzle, T. (2006). Ant colony optimization. IEEE computational intelligence magazine, 1(4), 28-39

    work page 2006

  8. [8]

    K., Huang, V

    Qin, A. K., Huang, V . L., & Suganthan, P. N. (2008). Differential evolution algorithm with strategy adaptation for global numerical op- timization. IEEE transactions on Evolutionary Computation, 13(2), 398- 417

    work page 2008

Show all 41 references
  1. [9]

    Tan, Y ., & Zhu, Y . (2010). Fireworks algorithm for optimization. In Advances in Swarm Intelligence: First International Conference, ICSI 2010, Beijing, China, June 12-15, 2010, Proceedings, Part I 1 (pp. 355- 364). Springer Berlin Heidelberg

    2010

  2. [10]

    (2021, June)

    Chen, M., & Tan, Y . (2021, June). Exponentially decaying explosion in fireworks algorithm. In 2021 IEEE Congress on Evolutionary Compu- tation (CEC) (pp. 1406-1413). IEEE

    2021

  3. [11]

    Li, J., & Tan, Y . (2017). Loser-out tournament-based fireworks algorithm for multimodal function optimization. IEEE Transactions on Evolution- ary Computation, 22(5), 679-691

    2017

  4. [12]

    Meng, X., & Tan, Y . (2024). Multi-guiding spark fireworks algorithm: Solving multimodal functions by multiple guiding sparks in fireworks algorithm. Swarm and Evolutionary Computation, 85, 101458

    2024

  5. [13]

    Jin, F., Liu, Y ., & Tan, Y . (2024). Derivative-free optimization for low- 9 TABLE VII: Comparing on CEC2013, dim = 50 idx NBIPOPaCMA.meanNBIPOPaCMA.stdNIPOPaCMA.meanNIPOPaCMA.stdSHADE.mean SHADE.std MVMO.mean MVMO.std SPSO2011.meanSPSO2011.std TFW A.meanTFW A.std 1 0.000e+00...

    2024

  6. [14]

    J., Shen, Y ., Wallis, P., Allen-Zhu, Z., Li, Y ., Wang, S.,

    Hu, E. J., Shen, Y ., Wallis, P., Allen-Zhu, Z., Li, Y ., Wang, S., ... & Chen, W. (2021). Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685

    2021 arXiv

  7. [15]

    J., Qu, B

    Liang, J. J., Qu, B. Y ., Suganthan, P. N., & Hern ´andez-D´ıaz, A. G. (2013). Problem definitions and evaluation criteria for the CEC 2013 special session on real-parameter optimization. Computational Intelligence Laboratory, Zhengzhou University, Zhengzhou, China and Nanyang...

    2013

  8. [16]

    Wu, G., Mallipeddi, R., & Suganthan, P. N. (2017). Problem definitions and evaluation criteria for the CEC 2017 competition on constrained real-parameter optimization. National University of Defense Technol- ogy, Changsha, Hunan, PR China and Kyungpook National University, Dae...

    2017

  9. [17]

    (2014, July)

    Li, J., Zheng, S., & Tan, Y . (2014, July). Adaptive fireworks algorithm. In 2014 IEEE Congress on evolutionary computation (CEC) (pp. 3214- 3221). IEEE

    2014

  10. [18]

    (2014, July)

    Zheng, S., Janecek, A., Li, J., & Tan, Y . (2014, July). Dynamic search in fireworks algorithm. In 2014 IEEE Congress on evolutionary computation (CEC) (pp. 3222-3229). IEEE

    2014

  11. [19]

    Li, J., Zheng, S., & Tan, Y . (2016). The effect of information utilization: Introducing a novel guiding spark in the fireworks algorithm. IEEE Transactions on Evolutionary Computation, 21(1), 153-166

    2016

  12. [20]

    Li, Y ., & Tan, Y . (2022). Fireworks algorithm based on search space partition. International Journal of Intelligent Systems, 37(12), 10783- 10812

    2022

  13. [21]

    Chen, M., & Tan, Y . (2023). SF-FW A: A Self-Adaptive Fast Fireworks Algorithm for effective large-scale optimization. Swarm and Evolution- ary Computation, 80, 101314

    2023

  14. [22]

    (1996, May)

    Hansen, N., & Ostermeier, A. (1996, May). Adapting arbitrary normal mutation distributions in evolution strategies: The covariance matrix adaptation. In Proceedings of IEEE international conference on evo- lutionary computation (pp. 312-317). IEEE

    1996

  15. [23]

    Akimoto, Y ., Nagata, Y ., Ono, I., & Kobayashi, S. (2012). Theoretical foundation for CMA-ES from information geometry perspective. Algo- rithmica, 64, 698-716

    2012

  16. [24]

    (2013, June)

    Loshchilov, I. (2013, June). CMA-ES with restarts for solving CEC 2013 benchmark problems. In 2013 IEEE Congress on Evolutionary Computation (pp. 369-376). Ieee

    2013

  17. [25]

    (2013, June)

    Tanabe, R., & Fukunaga, A. (2013, June). Evaluating the performance of SHADE on CEC 2013 benchmark problems. In 2013 IEEE Congress on evolutionary computation (pp. 1952-1959). IEEE

    2013

  18. [26]

    L., & Erlich, I

    Rueda, J. L., & Erlich, I. (2013, June). Hybrid mean-variance mapping optimization for solving the IEEE-CEC 2013 competition problems. In 2013 IEEE Congress on Evolutionary Computation (pp. 1664-1671). IEEE

    2013

  19. [27]

    (2013, June)

    Zambrano-Bigiarini, M., Clerc, M., & Rojas, R. (2013, June). Standard 10 particle swarm optimisation 2011 at cec-2013: A baseline for future pso improvements. In 2013 IEEE congress on evolutionary computation (pp. 2337-2344). IEEE

    2013

  20. [28]

    K., & Singh, D

    Kumar, A., Misra, R. K., & Singh, D. (2017, June). Improving the local search capability of effective butterfly optimizer using covariance matrix adapted retreat phase. In 2017 IEEE congress on evolutionary computation (CEC) (pp. 1835-1842). IEEE

    2017

  21. [29]

    W., Hadi, A

    Mohamed, A. W., Hadi, A. A., Fattouh, A. M., & Jambi, K. M. (2017, June). LSHADE with semi-parameter adaptation hybrid with CMA-ES for solving CEC 2017 benchmark problems. In 2017 IEEE Congress on evolutionary computation (CEC) (pp. 145-152). IEEE

    2017

  22. [30]

    (2017, June)

    Biedrzycki, R. (2017, June). A version of IPOP-CMA-ES algorithm with midpoint for CEC 2017 single objective bound constrained problems. In 2017 IEEE Congress on Evolutionary Computation (CEC) (pp. 1489- 1494). IEEE

    2017

  23. [31]

    I., & Douglas, S

    Amari, S. I., & Douglas, S. C. (1998, May). Why natural gradient?. In Proceedings of the 1998 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP’98 (Cat. No. 98CH36181) (V ol. 2, pp. 1213-1216). IEEE

    1998

  24. [32]

    (2011, July)

    Schaul, T., Glasmachers, T., & Schmidhuber, J. (2011, July). High di- mensions and heavy tails for natural evolution strategies. In Proceedings of the 13th annual conference on Genetic and evolutionary computation (pp. 845-852)

    2011

  25. [33]

    Rudolph, G. (1997). Local convergence rates of simple evolutionary algorithms with Cauchy mutations. IEEE Transactions on Evolutionary Computation, 1(4), 249-258

    1997

  26. [34]

    Ali, M., & Pant, M. (2011). Improving the performance of differential evolution algorithm using Cauchy mutation. Soft Computing, 15, 991- 1007

    2011

  27. [35]

    Kappler, C. (1996). Are evolutionary algorithms improved by large mu- tations?. In Parallel Problem Solving from Nature—PPSN IV: Interna- tional Conference on Evolutionary Computation—The 4th International Conference on Parallel Problem Solving from Nature Berlin, Germany, Sep...

    1996

  28. [36]

    J., Ahn, C

    Choi, T. J., Ahn, C. W., & An, J. (2013). An adaptive Cauchy differential evolution algorithm for global numerical optimization. The Scientific World Journal, 2013

    2013

  29. [37]

    L., Nguyen, D., & Nguyen, D

    Kirkby, J. L., Nguyen, D., & Nguyen, D. (2019). Moments of Student’s t-distribution: A unified approach. arXiv preprint arXiv:1912.01607

    2019 arXiv

  30. [38]

    Roth, M. (2012). On the multivariate t distribution. Link ¨oping University Electronic Press

    2012

  31. [39]

    G., & Joarder, A

    Kibria, B. G., & Joarder, A. H. (2006). A short review of multivariate t-distribution. Journal of Statistical research, 40(1), 59-72

    2006

  32. [40]

    Kotz, S., & Nadarajah, S. (2004). Multivariate t-distributions and their applications. Cambridge University Press

    2004

  33. [41]

    Akimoto, Y ., Nagata, Y ., Ono, I., & Kobayashi, S. (2012). Theoretical foundation for CMA-ES from information geometry perspective. Algo- rithmica, 64, 698-716. 11 APPENDIX We will give the derivation of the equation 17-19 p(x) = Γ( v 2 + d 2 ) Γ( v 2 ) ∗ |Σ|− 1 2 (v∗π) d 2 ∗...

    2012

Pith tools

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