Pith. sign in

REVIEW 4 major objections 4 minor 24 references

Energy and Quality of Surrogate-Assisted Search Algorithms: a First Analysis

T0 review · 4 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper tries to establish that surrogate-assisted optimization should be judged by energy and surrogate accuracy, not just runtime and solution quality, and reports a first measurement study of particle swarm optimization (PSO) versions

desk verdict A plausible first step on energy profiling for surrogate-assisted PSO, but the supplied text is unreadable—so the evidence can't be checked, and that's the whole ballgame. read the letter →

arxiv 2508.07691 v1 pith:HAS4RXXQ submitted 2025-08-11 cs.NE

Tomohiro Harada , Enrique Alba , Gabriel Luque This is my paper
classification cs.NE
keywords surrogate-assistedoptimizationparticleswarmneuralnetworksurrogatesenergyprofilingprocessorandmemorygateaccuracymetaheuristics
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 energy profile and surrogate accuracy are understudied but important dimensions for evaluating surrogate-assisted metaheuristics. It reports a first measurement study comparing particle swarm optimization variants, including versions with pre-trained and retrained neural-network surrogates, tracking both processor and memory energy. The authors position this as a first step toward a methodology that characterizes such algorithms by energy and surrogate accuracy in addition to time and numerical efficiency. If the paper is right, energy should become a standard reported metric when comparing optimization algorithms, and surrogate accuracy should be evaluated by its ability to drive search toward acceptable solutions, not just fit the training data.

What carries the argument

The central objects are PSO variants: standard PSO, PSO with a pre-trained neural network as a fitness surrogate, and PSO with a retrained or continuously updated neural network surrogate. The evaluation machinery is the joint measurement of processor energy, memory energy, and surrogate accuracy, used to compare the variants' search behavior and output quality.

What would settle it

Repeat the same benchmark suite and PSO variants on a different processor and memory configuration, keeping everything else fixed, and check whether the ordering of variants by processor and memory energy remains the same; if the ranking flips, the central energy comparison is not generalizable across hardware.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that surrogate-assisted PSO versions can be measurably distinguished by their energy consumption in the processor and memory, and that the accuracy of the neural-network surrogate meaningfully affects the search's ability to reach an acceptable solution. The authors treat these measurements as shedding new light on surrogate-assisted search and as the beginning of a more holistic assessment framework for optimization and learning techniques.

Load-bearing premise

The energy measurements are stable and general enough that the reported differences between PSO versions come from the algorithms themselves, not from the specific hardware, operating system, or measurement setup used in the experiments.

Editorial extensions

If this is right

  • Energy consumption should be reported alongside runtime and solution quality when comparing optimization algorithms.
  • Pre-trained and retrained surrogate models may exhibit different energy-accuracy tradeoffs, so the choice of surrogate training strategy becomes an energy-relevant design decision.
  • The proposed measurement approach can be applied to other surrogate-assisted metaheuristics, not just PSO.
  • Surrogate accuracy, measured by the ability to guide search to acceptable solutions, becomes a more meaningful quality metric than raw prediction error.
  • A first-step methodology emerges for holistic characterization of optimization techniques, covering time, numerical efficiency, energy, and surrogate accuracy.

Reading between the lines

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

  • The energy cost of training the surrogate model itself is likely separate from the energy of using it during search; a full accounting that amortizes training energy could change the ranking between pre-trained and retrained variants.
  • The findings suggest a broader hypothesis: that energy profiles are algorithm-specific and can be used as a selection criterion for green or energy-constrained computing environments.
  • A natural extension would be to test whether the energy-accuracy ordering observed here generalizes across different hardware, operating systems, or problem classes; if not, the methodology still stands but the specific rankings may be hardware-dependent.
  • The 'acceptable solution' yardstick could be standardized (e.g., a fixed threshold relative to known optima) so that surrogate accuracy is comparable across studies.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The manuscript proposes a first-step empirical study of the energy consumption of surrogate-assisted particle swarm optimization. It compares PSO variants using pre-trained and retrained neural-network surrogates, measuring processor and memory energy separately, and it assesses surrogate accuracy with respect to an 'acceptable solution' criterion. The authors argue that energy and surrogate accuracy should be considered alongside runtime and numerical quality. The abstract is readable and motivates the study, but the body of the paper is presented as unreadable mojibake in the provided submission, so the experimental setup, tables, and results could not be verified.

Significance. If the measurements are sound, this is a useful and timely contribution: energy profiling of surrogate-assisted metaheuristics, especially the separate treatment of processor and memory energy, is genuinely understudied. The proposed multidimensional assessment (quality, runtime, energy, surrogate accuracy) is a reasonable step for the community. However, the contribution is empirical, and no code, reproducible scripts, or machine-checked proofs are visible; the evaluation rests entirely on the experimental section, which is currently inaccessible.

major comments (4)
  1. [Full text (as rendered)] The body of the paper is presented as mojibake; Sections containing the experimental setup, algorithms, tables, and results are not legible. The central claim is an empirical comparison of energy and accuracy, so the measurement instrumentation, benchmark definitions, surrogate training protocol, repetition counts, and result tables cannot be checked. This is load-bearing and must be corrected before the manuscript can be evaluated.
  2. [Abstract and visible table fragments] Even from the legible parts, no experimental reporting details are available: hardware platform, energy measurement method (RAPL, external power meter, OS counters, or other), sampling rate, number of runs, variance or confidence intervals, or statistical testing. Without these, the claimed differences between PSO versions are not supported, and the paper currently reads as a qualitative research agenda rather than a measurement study.
  3. [Abstract ('acceptable solution')] The yardstick for surrogate accuracy is undefined. What counts as an 'acceptable solution' (absolute error threshold, success rate, fixed budget, or something else) is essential for interpreting the energy-versus-accuracy trade-off. The manuscript should define this criterion explicitly per benchmark and report how many runs meet it.
  4. [Surrogate training/evaluation protocol] The manuscript does not state whether surrogate accuracy is evaluated on training data, validation data, or held-out test data, nor is the retraining schedule described. If accuracy is measured on the data used to train the surrogate, the comparison is uninterpretable. This distinction is central to the claimed 'surrogate accuracy to properly drive the search' and must be clarified.
minor comments (4)
  1. [Abstract] The phrase 'shed new light' is vague. The abstract should state at least one concrete quantitative finding, such as an observed energy difference or an accuracy level, to give readers a basis for judging the contribution.
  2. [All tables] The visible table fragments lack labels, units, and sample sizes. Tables should include units (joules, watts), standard deviations or confidence intervals, and the number of repetitions per configuration.
  3. [Introduction/Conclusion] The claimed 'first step toward a methodology' would be better supported by an explicit description of the proposed methodology (variables measured, normalization, reporting format) rather than a general appeal to holistic assessment.
  4. [Formatting] The submitted full text is unreadable due to encoding corruption. Authors should verify the PDF/TeX encoding before resubmission, as this currently prevents any substantive review of the experiments.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity is visible; the paper is an empirical measurement study and no derivation step can be shown to reduce to its own inputs.

full rationale

The available readable text (abstract and some fragmentary sections) presents an empirical, measurement-oriented study: it compares PSO variants with and without surrogate neural networks in terms of processor and memory energy, and also examines surrogate accuracy. There is no derivation chain whose output is equivalent to an input by construction. I cannot identify any equation, fitted parameter, or self-citation that is load-bearing in a way that would make the claimed energy or accuracy results forced by definition. The strongest possible concern—that surrogate accuracy might be evaluated on training data—is not supported by any quoted text in the provided manuscript, and the full text is mostly corrupted, preventing any concrete reduction from being exhibited. Because the instructions require quoting specific evidence for circularity and forbid speculation, the appropriate finding is 'no significant circularity.'

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

No free parameters or invented entities are identifiable from the abstract alone; the PSO configuration, NN hyperparameters, and energy-measurement thresholds would live in the unreadable full text. The two domain assumptions listed are the minimum the central claim rests on beyond the experimental apparatus itself.

assumptions (2)
  • domain assumption A neural-network surrogate, pre-trained or retrained, yields an approximation of the true fitness function that is good enough to preserve the search behavior being measured.
    The abstract's whole setup replaces real fitness evaluations with NN outputs; this is standard in surrogate-assisted optimization but assumed, not derived.
  • domain assumption The energy-measurement procedure (hardware, counters, isolation, repetitions) yields representative processor and memory energy figures.
    The quantitative novelty of the paper is the energy profile, so the measuring apparatus is load-bearing; the abstract does not describe it.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Energy and Quality of Surrogate-Assisted Search Algorithms: a First Analysis." pith.science (2026). https://pith.science/paper/HAS4RXXQ

@misc{pith2026250807691,
  author       = {Pith},
  title        = {Pith review of: Energy and Quality of Surrogate-Assisted Search Algorithms: a First Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HAS4RXXQ}},
  note         = {Machine review of arXiv:2508.07691}
}
read the original abstract

Solving complex real problems often demands advanced algorithms, and then continuous improvements in the internal operations of a search technique are needed. Hybrid algorithms, parallel techniques, theoretical advances, and much more are needed to transform a general search algorithm into an efficient, useful one in practice. In this paper, we study how surrogates are helping metaheuristics from an important and understudied point of view: their energy profile. Even if surrogates are a great idea for substituting a time-demanding complex fitness function, the energy profile, general efficiency, and accuracy of the resulting surrogate-assisted metaheuristic still need considerable research. In this work, we make a first step in analyzing particle swarm optimization in different versions (including pre-trained and retrained neural networks as surrogates) for its energy profile (for both processor and memory), plus a further study on the surrogate accuracy to properly drive the search towards an acceptable solution. Our conclusions shed new light on this topic and could be understood as the first step towards a methodology for assessing surrogate-assisted algorithms not only accounting for time or numerical efficiency but also for energy and surrogate accuracy for a better, more holistic characterization of optimization and learning techniques.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

24 extracted references · 22 canonical work pages

  1. [1]

    Osaba, E

    E. Osaba, E. Villar-Rodriguez, J. Del Ser , A. J. Nebro, D. Molina, A. LaTorre, P. N. Suganthan, C. A. Coello Coello , and F. Herrera, ``A tutorial on the design, experimentation and application of metaheuristic algorithms to real-world optimization problems,'' Swarm and Evolutionary Computation, vol. 64, p. 100888, 2021

    work page 2021

  2. [2]

    Alba and J

    E. Alba and J. M. Troya, ``Improving flexibility and efficiency by adding parallelism to genetic algorithms,'' Statistics and Computing, vol. 12, no. 2, pp. 91--114, 2002

    work page 2002

  3. [3]

    Abdelhafez, G

    A. Abdelhafez, G. Luque, and E. Alba, ``Parallel execution combinatorics with metaheuristics: Comparative study,'' Swarm and Evolutionary Computation, vol. 55, p. 100692, 2020

    work page 2020

  4. [4]

    Khalfi, F

    S. Khalfi, F. Caraffini, and G. Iacca, ``Metaheuristics in the balance: A survey on memory-saving approaches for platforms with seriously limited resources,'' International Journal of Intelligent Systems, vol. 2023, p. 5708085, Nov 2023

    work page 2023

  5. [5]

    He and G

    J. He and G. Lin, ``Average convergence rate of evolutionary algorithms,'' IEEE Transactions on Evolutionary Computation, vol. 20, no. 2, pp. 316--321, 2016

    work page 2016

  6. [6]

    A. H. Halim, I. Ismail, and S. Das, ``Performance assessment of the metaheuristic optimization algorithms: an exhaustive review,'' Artificial Intelligence Review, vol. 54, no. 3, pp. 2323--2409, Mar 2021

    work page 2021

  7. [7]

    Abdelhafez, E

    A. Abdelhafez, E. Alba, and G. Luque, ``A component-based study of energy consumption for sequential and parallel genetic algorithms,'' The Journal of Supercomputing, vol. 75, no. 10, pp. 6194--6219, Oct 2019

    work page 2019

  8. [8]

    M. N. Jamil and A.-L. Kor, ``Analyzing energy consumption of nature-inspired optimization algorithms,'' Green Technology, Resilience, and Sustainability, vol. 2, no. 1, p. 1, Jan 2022

    work page 2022

Show all 24 references
  1. [9]

    S. G. Paul, A. Saha, M. S. Arefin, T. Bhuiyan, A. A. Biswas, A. W. Reza, N. M. Alotaibi, S. A. Alyami, and M. A. Moni, ``A comprehensive review of green computing: Past, present, and future research,'' IEEE Access, vol. 11, pp. 87\,445--87\,494, 2023

    2023

  2. [10]

    Iima and Y

    H. Iima and Y. Hazama, ``Genetic algorithm introducing the estimation of optimal objective function values in subproblems by pre-training,'' Neural Computing and Applications, pp. 1--18, 2023

    2023

  3. [11]

    J. Li, Q. Zhou, H. Williams, H. Xu, and C. Du, ``Cyber-physical data fusion in surrogate-assisted strength pareto evolutionary algorithm for PHEV energy management optimization,'' IEEE Transactions on Industrial Informatics, vol. 18, no. 6, pp. 4107--4117, 2022

    2022

  4. [12]

    H. Tong, C. Huang, L. L. Minku, and X. Yao, ``Surrogate models in evolutionary single-objective optimization: A new taxonomy and experimental study,'' Information Sciences, vol. 562, pp. 414--437, 2021

    2021

  5. [13]

    Verdecchia, J

    R. Verdecchia, J. Sallou, and L. Cruz, ``A systematic review of green AI ,'' WIREs Data Mining and Knowl. Disc., vol. 13, no. 4, p. e1507, 2023

    2023

  6. [14]

    Díaz-Álvarez, P

    J. Díaz-Álvarez, P. A. Castillo, F. F. de Vega, F. Chávez, and J. Alvarado, ``Population size influence on the energy consumption of genetic programming,'' Meas. and Control, vol. 55, no. 1-2, pp. 102--115, 2022

    2022

  7. [15]

    Fern \'a ndez de Vega, J

    F. Fern \'a ndez de Vega, J. D \'i az, J. \'A . Garc \'i a, F. Ch \'a vez, and J. Alvarado, ``Looking for energy efficient genetic algorithms,'' in Artificial Evolution. 1em plus 0.5em minus 0.4em Springer, 2020, pp. 96--109

    2020

  8. [16]

    Jin, ``Surrogate-assisted evolutionary computation: Recent advances and future challenges,'' Swarm and EC, vol

    Y. Jin, ``Surrogate-assisted evolutionary computation: Recent advances and future challenges,'' Swarm and EC, vol. 1, no. 2, pp. 61--70, 2011

    2011

  9. [17]

    C. He, Y. Zhang, D. Gong, and X. Ji, ``A review of surrogate-assisted evolutionary algorithms for expensive optimization problems,'' Expert Systems with Applications, vol. 217, p. 119495, 2023

    2023

  10. [18]

    M. Cui, L. Li, M. Zhou, and A. Abusorrah, ``Surrogate-assisted autoencoder-embedded evolutionary optimization algorithm to solve high-dimensional expensive problems,'' IEEE Transactions on Evolutionary Computation, vol. 26, no. 4, pp. 676--689, 2022

    2022

  11. [19]

    Wei, W.-N

    F.-F. Wei, W.-N. Chen, W. Mao, X.-M. Hu, and J. Zhang, ``An efficient two-stage surrogate-assisted differential evolution for expensive inequality constrained optimization,'' IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 53, no. 12, pp. 7769--7782, 2023

    2023

  12. [20]

    García-Nieto, E

    J. García-Nieto, E. Alba, and A. Carolina Olivera , ``Swarm intelligence for traffic light scheduling: Application to real urban areas,'' Engineering Applications of Artificial Intelligence, vol. 25, no. 2, pp. 274--283, 2012

    2012

  13. [21]

    David, E

    H. David, E. Gorbatov, U. R. Hanebutte, R. Khanna, and C. Le, `` RAPL : Memory power estimation and capping,'' in 2010 International Symposium on Low-Power Electronics and Design, 2010, pp. 189--194

    2010

  14. [22]

    Ferro, G

    M. Ferro, G. D. Silva, F. B. de Paula, V. Vieira, and B. Schulze, ``Towards a sustainable AI : A case study of energy efficiency in decision tree algorithms,'' Concurrency and Comp.: Pract. and Exp., vol. 35, 8 2023

    2023

  15. [23]

    L. F. W. Anthony, B. Kanding, and R. Selvan, ``Carbontracker: Tracking and predicting the carbon footprint of training deep learning models,'' ICML Workshop on Challenges in Deploying and monitoring Machine Learning Systems, July 2020, arXiv:2007.03051

    2020 arXiv

  16. [24]

    - ,z/􎾐 )(:/bЁ

    11em plus .33em minus .07em 4000 4000 100 4000 4000 500 `\.=1000 = #1 \@IEEEnotcompsoconly \@IEEEcompsoconly #1 * [1] 0pt [0pt][0pt] #1 * [1] 0pt [0pt][0pt] #1 * \| ** #1 \@IEEEauthorblockNstyle \@IEEEcompsocnotconfonly \@IEEEauthorblockAstyle \@IEEEcompsocnotconfonly \@IEEEco...

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.