Pith. sign in

REVIEW 2 major objections 1 minor 13 references

WASHH uses a whale optimization backbone with dynamic selection among multiple search behaviors and anchor refinement to outperform fixed algorithms on benchmarks and SVC tuning.

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

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

T0 review · grok-4.3

2026-06-30 21:15 UTC pith:6UPQPWOC

load-bearing objection WASHH adds an online reward selector to WOA plus a few standard operators and anchors, but the abstract gives no equations or tuning details for the controller, so the reported rank of 1.10 is hard to trust as unbiased. the 2 major comments →

arxiv 2605.28844 v1 pith:6UPQPWOC submitted 2026-05-13 cs.NE cs.LG

WASHH: An Anchor-Aware Whale-Guided Selection Hyper-Heuristic for Continuous Optimization and SVC Configuration

Yifu Zhao , Xiaofan Zou , Junhao Wei , Yanxiao Li , Baili Lu , Zhenhong Peng , Dexing Yao , Haochen Li
show 4 more authors
Qinbin He Sio-Kei Im Xu Yang Yapeng Wang
This is my paper
classification cs.NE cs.LG
keywords hyper-heuristicwhale optimizationcontinuous optimizationsupport vector classifierblack-box optimizationanchor refinementonline reward controllerbenchmark functions
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper establishes that committing to a single metaheuristic is unreliable under small evaluation budgets in black-box continuous optimization, so WASHH dynamically selects among several search behaviors instead. It keeps whale optimization as the main exploitation structure but adds PSO-style memory, GWO-style averaging, DE-style variation, local search, and anchor refinement as options. An online reward controller shifts evaluations toward behaviors that show improvements, while anchor refinement uses inexpensive reference points such as box centers. Experiments on ten 30-dimensional functions with 12,000 evaluations each show WASHH securing the best average rank and improving over plain whale optimization on eight functions. The same approach also produces the lowest validation log loss when tuning a support vector classifier under a 300-evaluation limit.

Core claim

WASHH is an anchor-aware selection hyper-heuristic that employs whale optimization as its exploitation backbone while treating PSO-style memory, GWO-style leader averaging, DE-style variation, local coordinate search, and anchor-guided refinement as selectable behaviors. An online reward controller allocates evaluations according to observed improvements, and anchor refinement exploits inexpensive reference configurations such as box centers or default model settings without bypassing black-box evaluation.

What carries the argument

The online reward controller that allocates evaluations to different search behaviors based on observed improvements, together with anchor-guided refinement that uses cheap reference points.

Load-bearing premise

The online reward controller allocates evaluations according to observed improvements in a way that does not itself introduce bias when comparing performance across different functions, runs, and the SVC task.

What would settle it

Independent runs on the same ten 30-dimensional functions with 12,000 evaluations each that produce an average rank for WASHH worse than 1.10 or show it failing to improve over WOA on at least three functions.

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

If this is right

  • WASHH achieves the best average rank of 1.10 and is best or tied best on all ten benchmark functions.
  • It strictly improves over WOA on eight functions while tying at the numerical optimum on Rastrigin and Griewank.
  • WASHH records the lowest mean validation log loss among compared optimizers when configuring SVC hyperparameters under a 300-evaluation budget.

Where Pith is reading between the lines

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

  • If the reward controller generalizes without bias, the same selection logic could be applied to other black-box problems that require mixing exploitation from one algorithm with operators from others.
  • Anchor refinement using reference points may extend naturally to additional inexpensive sources such as problem-specific defaults beyond box centers.
  • The overall design indicates that lightweight hyper-heuristics can serve as an alternative to single-algorithm choices when evaluation budgets are tight.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 1 minor

Summary. The manuscript proposes WASHH, a selection hyper-heuristic that uses WOA as the primary exploitation mechanism while treating PSO-style memory, GWO-style leader averaging, DE-style variation, local coordinate search, and anchor-guided refinement as selectable operators. An online reward controller dynamically allocates a fixed evaluation budget according to observed improvements, and anchor refinement exploits inexpensive reference points (e.g., box centers) without bypassing black-box evaluations. On ten 30-dimensional benchmark functions (10 runs, 12 000 evaluations) WASHH reports the best average rank (1.10) and is best or tied-best on every function, strictly improving over plain WOA on eight functions while tying at the numerical optimum on Rastrigin and Griewank. The same framework is applied to SVC hyper-parameter configuration for breast-cancer diagnosis under a 300-evaluation budget and again reports the lowest mean validation log-loss.

Significance. If the performance claims are reproducible and the reward controller can be shown to be neutral with respect to landscape and budget allocation, the work would supply a lightweight, anchor-aware hyper-heuristic template that could be useful for low-budget continuous and model-configuration tasks. The manuscript does not, however, supply the equations, pseudocode, or hyper-parameters of the reward controller, nor any statistical tests or variance measures, so the central empirical claims cannot yet be verified.

major comments (2)
  1. [Abstract / §3] Abstract (and presumably §3–4): the central performance claims (rank 1.10, strict improvement on 8/10 functions) rest on the online reward controller’s dynamic allocation of the fixed 12 000-evaluation budget. No equations, update rule, or hyper-parameters for the controller are supplied, making it impossible to determine whether the mechanism confers an advantage simply by concentrating evaluations on whichever operator improves fastest on a given run. This directly undermines the fairness of the comparison with fixed-allocation baselines such as WOA.
  2. [Abstract] Abstract: the reported results give only mean ranks and “best or tied-best” statements; no standard deviations, statistical tests (Wilcoxon, Friedman, etc.), or raw data tables are mentioned. Without these, the claim that WASHH “obtains the lowest mean validation log loss” on the SVC task cannot be assessed for significance.
minor comments (1)
  1. [Abstract] The acronym “LEAD systems” appears in the final sentence of the abstract but is never expanded or referenced elsewhere; either define it or remove the phrase.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on reproducibility and statistical validation. We address each major comment below and will revise the manuscript to incorporate the requested details.

read point-by-point responses

  1. Referee: [Abstract / §3] Abstract (and presumably §3–4): the central performance claims (rank 1.10, strict improvement on 8/10 functions) rest on the online reward controller’s dynamic allocation of the fixed 12 000-evaluation budget. No equations, update rule, or hyper-parameters for the controller are supplied, making it impossible to determine whether the mechanism confers an advantage simply by concentrating evaluations on whichever operator improves fastest on a given run. This directly undermines the fairness of the comparison with fixed-allocation baselines such as WOA.

    Authors: We agree that the reward controller description is insufficient for reproducibility. In the revised manuscript we will add the complete equations governing reward accumulation, the precise update rule for dynamic budget allocation, and all hyper-parameters (including learning rates and thresholds). This will enable direct verification that allocation is driven by observed per-operator improvement rather than arbitrary favoritism. revision: yes

  2. Referee: [Abstract] Abstract: the reported results give only mean ranks and “best or tied-best” statements; no standard deviations, statistical tests (Wilcoxon, Friedman, etc.), or raw data tables are mentioned. Without these, the claim that WASHH “obtains the lowest mean validation log loss” on the SVC task cannot be assessed for significance.

    Authors: We concur that variance measures and formal statistical tests are required. The revision will report standard deviations across the ten runs for all benchmarks and the SVC task, include Wilcoxon signed-rank tests for pairwise comparisons against WOA and other baselines, and apply the Friedman test with Nemenyi post-hoc analysis. Raw result tables will be supplied in supplementary material. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical results on fixed benchmarks

full rationale

The paper proposes WASHH, a selection hyper-heuristic, and reports its measured performance (rank 1.10, improvements over WOA) on ten standard 30D benchmark functions and an SVC task under fixed budgets. These are direct experimental outcomes of executing the full algorithm, not quantities derived by equations from parameters fitted to the same runs. No self-definitional steps, fitted-input predictions, or load-bearing self-citations appear in the abstract or description. The reward controller is part of the method under test; its effect is evaluated comparatively rather than assumed or reduced by construction. The derivation chain is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract supplies no equations, parameters, or background assumptions; ledger therefore empty.

pith-pipeline@v0.9.1-grok · 5795 in / 1163 out tokens · 32817 ms · 2026-06-30T21:15:06.790380+00:00 · methodology

0 comments
read the original abstract

Learning-assisted algorithm design often has to make reliable search decisions under small evaluation budgets, where committing to a single metaheuristic can be unreliable. We propose WASHH, a Whale-guided Adaptive Selection Hyper-Heuristic for continuous black-box optimization. WASHH uses WOA as the main exploitation backbone, but treats PSO-style memory, GWO-style leader averaging, DE-style variation, local coordinate search, and anchor-guided refinement as selectable search behaviors. An online reward controller allocates evaluations according to observed improvements, while anchor refinement exploits inexpensive reference configurations such as box centers or default model settings without bypassing black-box evaluation. On ten 30-dimensional benchmark functions with 10 independent runs and 12,000 evaluations, WASHH achieves the best average rank, 1.10, and is best or tied best on all ten functions. It strictly improves over WOA on eight functions and ties WOA at the numerical optimum on Rastrigin and Griewank. We further study SVC hyperparameter configuration for breast cancer diagnosis under a 300-evaluation budget. WASHH obtains the lowest mean validation log loss among the compared optimizers, suggesting that anchor-aware selection hyper-heuristics are a practical lightweight direction for LEAD systems.

Figures

Figures reproduced from arXiv: 2605.28844 by Baili Lu, Dexing Yao, Haochen Li, Junhao Wei, Qinbin He, Sio-Kei Im, Xiaofan Zou, Xu Yang, Yanxiao Li, Yapeng Wang, Yifu Zhao, Zhenhong Peng.

Figure 1
Figure 1. Figure 1: Schematic overview of breast cancer diagnosis model configuration with WASHH. Candidate SVC hyperparameters are [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Mean best-so-far convergence curves over 10 independent runs for all compared algorithms on the ten benchmark func [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Breast cancer diagnosis model configuration. The curves show mean validation log loss during the HPO search, with [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Final validation-set diagnostic performance of the SVC configured by WASHH, shown by the ROC curve and confusion [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

13 extracted references · 13 canonical work pages

  1. [1]

    Burke, Michel Gendreau, Matthew Hyde, Graham Kendall, Gabriela Ochoa, Ender Ozcan, and Rong Qu

    [Burkeet al., 2013] Edmund K. Burke, Michel Gendreau, Matthew Hyde, Graham Kendall, Gabriela Ochoa, Ender Ozcan, and Rong Qu. Hyper-heuristics: A survey of the state of the art.Journal of the Operational Research Society, 64(12):1695–1724,

    work page 2013

  2. [2]

    A hyperheuristic approach to scheduling a sales summit

    [Cowlinget al., 2001] Peter Cowling, Graham Kendall, and Eric Soubeiga. A hyperheuristic approach to scheduling a sales summit. InPractice and Theory of Automated Timetabling III, volume 2079 ofLecture Notes in Computer Science, pages 176–190. Springer,

    work page 2001

  3. [3]

    Analyzing bandit-based adaptive operator selection mechanisms.Annals of Mathematics and Artificial Intelligence, 60(1–2):25–64,

    [Fialhoet al., 2010] Alvaro Fialho, Luis Da Costa, Marc Schoenauer, and Michele Sebag. Analyzing bandit-based adaptive operator selection mechanisms.Annals of Mathematics and Artificial Intelligence, 60(1–2):25–64,

    work page 2010

  4. [4]

    GWOA: A multi- strategy enhanced whale optimization algorithm for engineering design optimization.PLOS ONE, 20(9):e0322494,

    [Guet al., 2025] Yanzhao Gu, Junhao Wei, Zikun Li, Baili Lu, Shirou Pan, and Ngai Cheong. GWOA: A multi- strategy enhanced whale optimization algorithm for engineering design optimization.PLOS ONE, 20(9):e0322494,

    work page 2025

  5. [5]

    Particle swarm optimization

    [Kennedy and Eberhart, 1995] James Kennedy and Russell Eberhart. Particle swarm optimization. InProceedings of the IEEE International Conference on Neural Networks, pages 1942–1948, Perth, Australia,

    work page 1995

  6. [6]

    [Liet al., 2026] Zikun Li, Wenxuan Zhu, Ran Zhang, et al

    IEEE. [Liet al., 2026] Zikun Li, Wenxuan Zhu, Ran Zhang, et al. ASKSSA-CNN-BiLSTM: A novel time series forecasting model for stock price prediction based on an enhanced sparrow search algorithm. InProceedings of the 6th Asia Conference on Information Engineering, pages 20–26, Nanyang Technological University, Singapore,

    work page 2026

  7. [7]

    The whale optimization algorithm.Advances in Engineering Software, 95:51–67,

    [Mirjalili and Lewis, 2016] Seyedali Mirjalili and Andrew Lewis. The whale optimization algorithm.Advances in Engineering Software, 95:51–67,

    work page 2016

  8. [8]

    [Mirjaliliet al., 2014] Seyedali Mirjalili, Seyed Mohammad Mirjalili, and Andrew Lewis

    8 0.0 0.2 0.4 0.6 0.8 1.0 False positive rate 0.0 0.2 0.4 0.6 0.8 1.0True positive rate ROC, AUC=0.997 malignant benign Predicted label malignant benign True label 73 2 4 121 Validation confusion matrix Figure 4: Final validation-set diagnostic performance of the SVC configured by W ASHH, shown by the ROC curve and confusion matrix. [Mirjaliliet al., 2014...

    work page 2014

  9. [9]

    Scikit-learn: Machine learning in python.Journal of Machine Learning Research, 12:2825–2830,

    [Pedregosaet al., 2011] Fabian Pedregosa, Gael Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Weiss, Vincent Dubourg, Jake Vanderplas, Alexandre Passos, David Cournapeau, Matthieu Brucher, Matthieu Perrot, and Edouard Duchesnay. Scikit-learn: Machine learning in python.Journal of M...

    work page 2011

  10. [10]

    Differential evolution–a simple and efficient heuristic for global optimization over continuous spaces.Journal of Global Optimization, 11(4):341–359,

    [Storn and Price, 1997] Rainer Storn and Kenneth Price. Differential evolution–a simple and efficient heuristic for global optimization over continuous spaces.Journal of Global Optimization, 11(4):341–359,

    work page 1997

  11. [11]

    Nick Street, William H

    [Streetet al., 1993] W. Nick Street, William H. Wolberg, and Olvi L. Mangasarian. Nuclear feature extraction for breast tumor diagnosis. InProceedings of the IS&T/SPIE International Symposium on Electronic Imaging: Science and Technology, pages 861–870,

    work page 1993

  12. [12]

    [Weiet al., 2024] Junhao Wei, Yanzhao Gu, K. L. Eddie Law, and Ngai Cheong. Adaptive position updating particle swarm optimization for uav path planning. InProceedings of the 22nd International Symposium on Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks, pages 124–131. IEEE,

    work page 2024

  13. [13]

    TSWOA: An enhanced WOA with triangular walk and spiral flight for engineering design optimization

    [Weiet al., 2025] Junhao Wei, Yanzhao Gu, Yuzheng Yan, Yapeng Wang, Zikun Li, Baili Lu, Shirou Pan, and Ngai Cheong. TSWOA: An enhanced WOA with triangular walk and spiral flight for engineering design optimization. InProceedings of the 8th International Conference on Advanced Algorithms and Control Engineering, pages 186–

    work page 2025