Pith. sign in

REVIEW 4 major objections 4 minor 58 references

Enhanced Ideal Objective Vector Estimation for Evolutionary Multi-Objective Optimization

T0 review · 4 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 The paper claims that the standard population-based way of estimating the ideal objective vector fails under distance-related, position-related, and mixed bias, and that actively solving m extreme weighted-sum subproblems with adaptive…

desk verdict A useful biased test suite and a plausible estimation component, but the headline result is inflated by a circular metric and an unverified normalization assumption. read the letter →

arxiv 2505.21903 v1 pith:E2DFNJ2Y submitted 2025-05-28 cs.NE

Ruihao Zheng , Zhenkun Wang , Yin Wu , Maoguo Gong This is my paper
classification cs.NE
keywords idealobjectivevectorevolutionarymulti-objectiveoptimizationbiasedtestproblemsextremeweightedsumpopulation-basedestimationnormalizationCMA-ESproblemgenerator
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 targets the ideal objective vector—the vector of each objective's best possible value—which many multi-objective evolutionary algorithms use as a reference for search guidance and normalization. It argues that the standard population-based estimate, taking the minimum objective value seen so far in the population, becomes unreliable exactly when the problem is biased: when random and evolved solutions cluster away from the Pareto front or away from its boundaries. To fix this, the paper proposes EIE, a plug-and-play component that actively searches m extreme weighted-sum subproblems with an adaptive, fine-grained optimizer, and feeds the resulting near-ideal solutions back into the host algorithm. The paper claims that replacing population-based estimation with EIE lowers the estimation error and improves Pareto-front approximation across three families of MOEAs on both its 16 newly designed biased test instances and 55 existing ones. A theorem bounds the per-component error by alpha_i beta/(1-alpha_i) under a normalized-range assumption, which is the theoretical basis for choosing the subproblem weights.

What carries the argument

The central object is the extreme weighted sum (EWS) subproblem: for objective i, minimize g_i(x|alpha_i) = (1-alpha_i) f_i(x) + (alpha_i/(m-1)) sum_{j != i} f_j(x) with a small positive alpha_i. This is a special case of weighted-sum scalarization designed so that its optimum is a Pareto-optimal point aligned with the ideal objective vector, rather than a weakly Pareto-optimal dominance-resistant point. Theorem 1 states that after normalizing all objective ranges to a common length beta, the error in the i-th component of the EWS solution is at most alpha_i beta/(1-alpha_i), which justifies setting alpha_i from a user-specified tolerance epsilon_i via alpha_i = epsilon_i/(epsilon_i+1). The companion machinery is the optimizer: one PSA-CMA-ES process per subproblem, which adaptively adjusts its population size and search covariance to explore a small biased region, with warm starting from the current population and injection of new solutions to avoid local optima.

What would settle it

Construct a two-objective biased problem where the true ideal and nadir values lie outside the range spanned by every population ever sampled, run EIE with epsilon_i = 0.05, and measure the normalized error on the i-th component. If the error exceeds alpha_i beta/(1-alpha_i) with beta set from the true range, the paper's Theorem 1 guarantee fails as stated; comparing EIE against a population-based estimator on this instance also tests whether active search actually recovers the ideal.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the ideal objective vector can be actively computed during an evolutionary run rather than passively read off from the population. Population-based estimators fail under distance-related, position-related, and mixed bias because the population never samples the extremes that define the ideal values. EIE instead decomposes the problem into m extreme weighted-sum subproblems, each tilting nearly all weight onto one objective; solving these subproblems yields m Pareto-optimal solutions whose objective values approximate the ideal vector. The method runs alongside the MOEA, adaptively consuming computational resources, and injects its solutions into the population so that the ideal-vector information directly guides the search. Experiments across dominance-based, decomposition-based, and indicator-based algorithms report that the estimation error drops to no more than about 5% on nearly all proposed instances, and that hypervolume improves accordingly.

Load-bearing premise

The whole scheme leans on the assumption that the objective space can be normalized with the true equal range for every objective, yet in the implementation that range is taken from the very population whose bias the method is meant to overcome.

Editorial extensions

If this is right

  • MOEAs can obtain a reliable ideal objective vector on biased problems without changing their selection or reproduction logic: EIE is a plug-and-play component.
  • Better ideal-vector estimates translate into better normalized objective spaces and reference points, and the experiments show higher hypervolume and faster anytime convergence on the biased test instances.
  • The m EWS subproblems produce Pareto-optimal boundary solutions, so EIE avoids the dominance-resistant solutions that arise when objectives are optimized separately.
  • Because EIE adaptively allocates function evaluations and terminates when its subproblems converge, integrating it does not impose a fixed overhead on the host MOEA.
  • The 16 test instances provide a controlled way to measure how distance-related, position-related, and mixed biases degrade ideal-vector estimation in future MOEAs.

Reading between the lines

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

  • The implementation normalizes with the current population's min/max, so a natural variant is to supply the normalizing range from an independent source and check whether the Theorem 1 error bound then holds exactly under strong bias; the paper does not test this variant.
  • Transferring EIE to discrete or mixed multi-objective problems would require replacing the continuous-space subproblem solver; the paper names discrete and mixed MOPs as future work.
  • The test generator's bias parameters could double as a diagnostic suite for measuring how much bias any MOEA tolerates before its reference-point estimation degrades.
  • The appendix's epsilon_i trade-off suggests an adaptive tolerance schedule could preserve boundary coverage while keeping estimation errors small; the paper leaves this open.
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 studies how distance-, position-, and mixed-bias in multi-objective optimization problems degrade population-based estimation of the ideal objective vector. It proposes a test-problem generator with adjustable bias parameters, instantiates 16 biased test instances, and introduces a plug-and-play component called EIE that solves m extreme-weighted-sum (EWS) subproblems with PSA-CMA-ES and injects the resulting solutions into a host MOEA. The paper reports that EIE reduces the E metric (ideal-vector estimation error) and improves hypervolume on the proposed instances and on a subset of bbob-biobj problems, and it provides a theoretical bound in Theorem 1 on the EWS subproblem error under an equal-ranges normalization assumption.

Significance. If the claims are sustained, the work would be a useful contribution: it systematizes bias categories, provides tunable test instances, and proposes a general mechanism for steering MOEAs toward ideal-vector boundary regions. The manuscript also ships source code and includes experiments on three structurally different MOEAs and on 55 existing bbob-biobj problems. However, the central empirical evidence is weakened by a circularity in the E metric, and the theoretical guarantee as stated is not established by the Appendix A proof nor verified under the implementation's population-based normalization. These issues are load-bearing for the paper's main claim that EIE accurately estimates the ideal objective vector.

major comments (4)
  1. [Section V-A3 and Algorithm 1] The E metric in Eq. (22) is computed from the minima of the final population. EIE explicitly injects into the population the solutions produced by its EWS subproblem optimizers (Algorithm 1 line 5 and Algorithm 2), and those subproblems are exactly designed to minimize the components of E. Thus the large reductions in Table II are partly by construction: the method is being evaluated on the very quantity it directly optimizes. The paper should provide an independent assessment of ideal-vector estimation accuracy, for example by comparing the internally maintained estimated ideal vector (the one used for normalization) against the true ideal vector on the proposed instances, or by measuring the accuracy of the m solutions output by EIE without letting them enter the final population before the metric is computed.
  2. [Appendix A, proof of Theorem 1] The proof's central step is not valid as written. After Eq. (25), the sentence "If we only consider Z subset [0,1]^m, then Eq. (25) is equal to 1" is false in general: sup_Z min_{z in Z} g(z) equals 1 only if each Z contains the all-ones vector, which is not guaranteed by Z subset [0,1]^m. Moreover, Eq. (26) does not imply the specific vertex (z_i = 0, z_j = 1 for j != i) used to conclude c_i = alpha_i. Since c_i = alpha_i is the basis for the bound in Eq. (21), the proof is incomplete. The authors should either correct the proof under explicit conditions that ensure the relevant vertex lies in Z, or restate Theorem 1 with assumptions that are actually sufficient.
  3. [Section IV-B and Eq. (26)] Theorem 1 assumes the objective space is normalized so that z_nad_j = z_ide_j + beta for all j, and that the feasible objective region satisfies the attainability conditions in Eq. (26). The implementation, however, normalizes using the current population's minimal and maximal objective values, which Section II-B itself argues are exactly the quantities that are unreliable under bias. The manuscript never reports on MOP1-MOP16 or the bbob-biobj suite whether the population-based normalization errors are small enough for the theorem's assumptions to hold, nor whether Eq. (26) is satisfied on those instances. Without such verification, Theorem 1 does not provide a theoretical guarantee for the implemented EIE algorithm. The authors should measure the normalization error (estimated min/max versus true z_ide and z_nad) during the runs, or change the normalization to a method with verifiable accuracy, and then check the theorem's assumptions.
  4. [Section V-B and Table II] The claim that EIE improves ideal-vector estimation is also supported by comparing E values only against the original MOEAs' final populations. Since EIE is allocated a substantial share of the function evaluations (up to about 57% on some instances, Table IV), the comparison conflates the quality of the estimated ideal vector with the allocation of computational budget. A fairer test would compare the estimated ideal vectors produced by EIE against those produced by population-based methods under matched conditions, for example by reporting the error of the internally estimated vector at the same number of function evaluations, independent of how those evaluations are split between the MOEA and EIE.
minor comments (4)
  1. [Section II-A] The three bias categories (distance-related, position-related, and mixed) are introduced through figures and qualitative descriptions; a short formal definition, for instance in terms of the density of f(X) for uniformly sampled X, would make the taxonomy easier to apply to the generator and to future benchmarks.
  2. [Table I] The table formatting makes some columns run together in the text (for example, the entries for s, p, c_pos, and gamma appear as a single string such as "5(1,1)^T (0.1,0.9)^T 0.1"). The authors should format the table so that each parameter column is unambiguous.
  3. [Section IV-C] The sentence "integrating m PSA-CMA-ES procedures into the MOEA does not increase the overall computational costs" is misleading: Table IV shows that EIE consumes between 10% and 57% of the total function evaluations, so the host MOEA necessarily receives a smaller share of the fixed budget. The authors should rephrase this as a budget-allocation trade-off rather than a claim of zero cost.
  4. [Section V-A3] The E metric in Eq. (22) is a root-sum-of-squares of normalized deviations; the paper should specify whether the normalization in the metric uses the true z_ide and z_nad of each instance, and if so, how these values are obtained for the bbob-biobj suite, where the true nadir may not be readily available.

Circularity Check

1 steps flagged · score 6.0 of 10

E metric validation is partially circular: EIE injects the very near-ideal solutions on which the metric is measured; HV and bbob-biobj provide independent support.

  1. self definitional [Section V-A3 (Eq. 22), Algorithm 1 lines 5/8, Section IV-B (Eq. 20)]
    "Let P be the final population. z e is extracted by z e_i = min_{x∈P} f_i (x) for i = 1, . . . , m. Create an offspring set O 1 via EIE. Select the next generation of P from O 1 ∪ O 2 ∪ P. min g ews_i (x|α_i) = (1−α_i)f_i(x)+ α_i/(m−1) Σ_{j≠i} f_j(x)."

    The E metric measures the distance from the true ideal vector to the per-objective minima of the final population P. Algorithm 1 explicitly puts EIE's offspring set O_1 into the selection pool, so P contains the outputs of the m EWS subproblem searches. Each EWS subproblem places weight (1−α_i)≈1 on f_i and is solved by a dedicated PSA-CMA-ES procedure, so its solutions are by design points with very low f_i. Hence the large E reductions in Table II largely transcribe EIE's own optimized outputs into the measured population; the E-based demonstration that EIE improves ideal-vector estimation is close to a restatement of the method's definition. The independent content comes from the HV improvements and bbob-biobj results, which do not reduce to this construction.

full rationale

The strongest circular step is the E-metric validation: because z_e is computed on the final population into which EIE injects its own EWS-optimized solutions, the reported E reductions are expected by construction. This is an open part of the algorithm, not a hidden fit, but it means Table II does not independently test the central estimation claim. HV gains, the inverted-PF experiments, and bbob-biobj provide separate, non-circular evidence, so the paper is only partially circular. The self-citations ([18], [51]) are used for baselines and generator components rather than as load-bearing justification; Theorem 1 is proved in Appendix A. The mismatch between Theorem 1's normalization assumptions and the implementation's population-based normalization is a correctness risk, not a circularity. Overall score 6: one central experimental result reduces by construction, while other evidence remains independent.

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

The central claim rests on a user-set tolerance parameter and a set of hand-chosen test generation parameters, plus assumptions about the geometry of the objective space and the capability of the inner optimizer. No new physical or abstract entities are postulated.

free parameters (3)
  • epsilon_i (alpha_i) = 0.05 for all objectives
    User-set tolerance for the EWS subproblem. It is hand-chosen, affects the theoretical error bound, and is not fitted to data.
  • Position bias parameters (c_pos, gamma) = Varies per instance, see Table I
    Hand-chosen to create position-related bias in the proposed test instances. These are design choices, not fitted parameters.
  • Distance bias parameters (a1-a5, c_dis) = Varies per instance, see Table I
    Hand-chosen to create distance-related and mixed biases in the test suite. They control the difficulty and are not fitted to data.
assumptions (4)
  • standard math For positive weights, an optimal solution of a weighted sum subproblem is Pareto-optimal.
    Used implicitly in the EWS method, Section IV-B. All weights are positive when alpha_i > 0 and alpha_i <= 0.5.
  • domain assumption The feasible objective region satisfies the boundary conditions in Eq. (26): for each objective, there exist nondominated points attaining the ideal and nadir values.
    Required for the proof of Theorem 1 in Appendix A. This is not guaranteed for arbitrary biased MOPs.
  • domain assumption Objective functions have equal ranges after normalization: z_nad_j = z_ide_j + beta for all j.
    Stated in Theorem 1 and used in the proof. The implementation estimates these ranges from the current population, which is unreliable under bias.
  • domain assumption PSA-CMA-ES can solve the EWS subproblems to a sufficient accuracy within the allocated function evaluation budget.
    The experimental evidence depends on the continuous optimizer succeeding; the paper notes this is limited to continuous MOPs.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhanced Ideal Objective Vector Estimation for Evolutionary Multi-Objective Optimization." pith.science (2026). https://pith.science/paper/E2DFNJ2Y

@misc{pith2026250521903,
  author       = {Pith},
  title        = {Pith review of: Enhanced Ideal Objective Vector Estimation for Evolutionary Multi-Objective Optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/E2DFNJ2Y}},
  note         = {Machine review of arXiv:2505.21903}
}
abstract

The ideal objective vector, which comprises the optimal values of the $m$ objective functions in an $m$-objective optimization problem, is an important concept in evolutionary multi-objective optimization. Accurate estimation of this vector has consistently been a crucial task, as it is frequently used to guide the search process and normalize the objective space. Prevailing estimation methods all involve utilizing the best value concerning each objective function achieved by the individuals in the current or accumulated population. However, this paper reveals that the population-based estimation method can only work on simple problems but falls short on problems with substantial bias. The biases in multi-objective optimization problems can be divided into three categories, and an analysis is performed to illustrate how each category hinders the estimation of the ideal objective vector. Subsequently, a set of test instances is proposed to quantitatively evaluate the impact of various biases on the ideal objective vector estimation method. Beyond that, a plug-and-play component called enhanced ideal objective vector estimation (EIE) is introduced for multi-objective evolutionary algorithms (MOEAs). EIE features adaptive and fine-grained searches over $m$ subproblems defined by the extreme weighted sum method. EIE finally outputs $m$ solutions that can well approximate the ideal objective vector. In the experiments, EIE is integrated into three representative MOEAs. To demonstrate the wide applicability of EIE, algorithms are tested not only on the newly proposed test instances but also on existing ones. The results consistently show that EIE improves the ideal objective vector estimation and enhances the MOEA's performance.

Figures

Figures reproduced from arXiv: 2505.21903 by the authors.

Figure 1
Figure 1. Plots of 10000 random solutions. The P F of (b) is taken from [22]. The P F of (c) is obtained by the exact solver. solution set of Q and the current population [15], [16]. Some studies consider a more optimistic estimation [17], [18], i.e., z e i = z ′ i − βi for i = 1, . . . , m, (3) where z ′ i can be calculated by Eq. (2) and βi is a non-negative value. βi is also suggested to decrease as the number of generatio… view at source ↗
Figure 2
Figure 2. Plots of random solutions on a two-objective problem with different biases. Letting [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The comparison between the position-related bias and distance-related [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: Plots of the population-based estimated ideal objective vector on [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 8
Figure 8. Figure 8: Plot of the final population of NSGA-II [1] on BT3 [26]. The [PITH_FULL_IMAGE:figures/full_fig_p004_8.png]
Figure 6
Figure 6. Figure 6: Plots of the population-based estimated ideal objective vector on [PITH_FULL_IMAGE:figures/full_fig_p004_6.png]
Figure 9
Figure 9. Figure 9: Plot of xˆi with γ = 0.1 and cˆi = 0.25. Note that the extreme objective vector (e.g., (1, 0, . . . , 0)⊺ ) is obtained when xˆi = 0 or 1 for i = 1, . . . , m and the corre￾sponding distance function values are zero. σi = 0 and σi = 1 are both mapped to xˆi = ˆci , whi…
Figure 10
Figure 10. Figure 10: Illustrations of r in the 2-objective and 3-objective cases [PITH_FULL_IMAGE:figures/full_fig_p006_10.png]
Figure 11
Figure 11. Figure 11: Plots of 10000 random solutions on proposed MOPs. [PITH_FULL_IMAGE:figures/full_fig_p007_11.png]
Figure 12
Figure 12. Figure 12: Plots of the non-dominated sets with the median [PITH_FULL_IMAGE:figures/full_fig_p012_12.png]
Figure 13
Figure 13. Figure 13: Curves of the metric value versus the number of function evaluations with the median metric values obtained by the MOEA and the MOEA with [PITH_FULL_IMAGE:figures/full_fig_p013_13.png]
Figure 14
Figure 14. Figure 14: Illustration of the proof of Theorem 1. Proof. Since z nad j = z ide j + β for j = 1, . . . , m, we let z ide j = 0 for j = 1, . . . , m and β = 1 without loss of gener￾ality. Then z nad j = 1 for j = 1, . . . , m. Recall that z i is the optimal objective vector of th…
Figure 15
Figure 15. Figure 15: Plots of the non-dominated sets with the median [PITH_FULL_IMAGE:figures/full_fig_p022_15.png]
Figure 16
Figure 16. Figure 16: Plots of the non-dominated sets with the median [PITH_FULL_IMAGE:figures/full_fig_p023_16.png]
Figure 17
Figure 17. Figure 17: Plots of the non-dominated sets with the median [PITH_FULL_IMAGE:figures/full_fig_p024_17.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

58 extracted references · 52 canonical work pages

  1. [1]

    A fast and elitist multiobjective genetic algorithm: NSGA-II,

    K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, “A fast and elitist multiobjective genetic algorithm: NSGA-II,”IEEE Transactions on Evolutionary Computation, vol. 6, no. 2, pp. 182–197, 2002. 14

    work page 2002

  2. [2]

    Indicator-based selection in multiobjective search,

    E. Zitzler, S. K ¨unzliet al., “Indicator-based selection in multiobjective search,” inInternational Conference on Parallel Problem Solving from Nature (PPSN). Springer, 2004, pp. 832–842

    work page 2004

  3. [3]

    MOEA/D: A multiobjective evolutionary algorithm based on decomposition,

    Q. Zhang and H. Li, “MOEA/D: A multiobjective evolutionary algorithm based on decomposition,”IEEE Transactions on Evolutionary Compu- tation, vol. 11, no. 6, pp. 712–731, 2007

    2007

  4. [4]

    Bi-goal evolution for many-objective optimization problems,

    M. Li, S. Yang, and X. Liu, “Bi-goal evolution for many-objective optimization problems,”Artificial Intelligence, vol. 228, pp. 45–65, 2015

    work page 2015

  5. [5]

    Mathematical runtime analysis for the non- dominated sorting genetic algorithm II (NSGA-II),

    W. Zheng and B. Doerr, “Mathematical runtime analysis for the non- dominated sorting genetic algorithm II (NSGA-II),”Artificial Intelli- gence, vol. 325, p. 104016, 2023

    work page 2023

  6. [6]

    An evolutionary many-objective optimization algorithm using reference-point-based nondominated sorting approach, part I: Solving problems with box constraints,

    K. Deb and H. Jain, “An evolutionary many-objective optimization algorithm using reference-point-based nondominated sorting approach, part I: Solving problems with box constraints,”IEEE Transactions on Evolutionary Computation, vol. 18, no. 4, pp. 577–601, 2014

    work page 2014

  7. [7]

    An enhanced decomposition- based evolutionary algorithm with adaptive reference vectors,

    M. Asafuddoula, H. K. Singh, and T. Ray, “An enhanced decomposition- based evolutionary algorithm with adaptive reference vectors,”IEEE Transactions on Cybernetics, vol. 48, no. 8, pp. 2321–2334, 2018

    work page 2018

  8. [8]

    A many-objective evolutionary algorithm using a one-by-one selection strategy,

    Y . Liu, D. Gong, J. Sun, and Y . Jin, “A many-objective evolutionary algorithm using a one-by-one selection strategy,”IEEE Transactions on Cybernetics, vol. 47, no. 9, pp. 2689–2702, 2017

    work page 2017

Show all 58 references
  1. [9]

    Effects of including optimal solutions into initial population on evolutionary multi- objective optimization,

    C. Gong, Y . Nan, L. M. Pang, Q. Zhang, and H. Ishibuchi, “Effects of including optimal solutions into initial population on evolutionary multi- objective optimization,” inProceedings of the Genetic and Evolutionary Computation Conference (GECCO), 2023, pp. 661–669

    2023

  2. [10]

    A survey of normalization methods in multiobjective evolutionary algorithms,

    L. He, H. Ishibuchi, A. Trivedi, H. Wang, Y . Nan, and D. Srinivasan, “A survey of normalization methods in multiobjective evolutionary algorithms,”IEEE Transactions on Evolutionary Computation, vol. 25, no. 6, pp. 1028–1048, 2021

    2021

  3. [11]

    Effects of objective space normalization in multi-objective evolutionary algorithms on real- world problems,

    L. He, Y . Nan, H. Ishibuchi, and D. Srinivasan, “Effects of objective space normalization in multi-objective evolutionary algorithms on real- world problems,” inProceedings of the Genetic and Evolutionary Computation Conference (GECCO), 2023, pp. 670–678

    2023

  4. [12]

    Interrelationship-based selec- tion for decomposition multiobjective optimization,

    K. Li, S. Kwong, Q. Zhang, and K. Deb, “Interrelationship-based selec- tion for decomposition multiobjective optimization,”IEEE Transactions on Cybernetics, vol. 45, no. 10, pp. 2076–2088, 2015

    2015

  5. [13]

    On the effect of normalization in MOEA/D for multi-objective and many-objective optimization,

    H. Ishibuchi, K. Doi, and Y . Nojima, “On the effect of normalization in MOEA/D for multi-objective and many-objective optimization,”Com- plex & Intelligent Systems, vol. 3, no. 4, pp. 279–294, 2017

    2017

  6. [14]

    Adapting decomposed directions for evolutionary multiobjective optimization,

    Y . Su, Q. Lin, Z. Ming, and K. C. Tan, “Adapting decomposed directions for evolutionary multiobjective optimization,”IEEE Transactions on Cybernetics, vol. 53, no. 10, pp. 6289–6302, 2023

    2023

  7. [15]

    A constrained decomposition approach with grids for evolutionary multiobjective optimization,

    X. Cai, Z. Mei, Z. Fan, and Q. Zhang, “A constrained decomposition approach with grids for evolutionary multiobjective optimization,”IEEE Transactions on Evolutionary Computation, vol. 22, no. 4, pp. 564–577, 2018

    2018

  8. [16]

    A strengthened dominance relation considering convergence and diversity for evolution- ary many-objective optimization,

    Y . Tian, R. Cheng, X. Zhang, Y . Su, and Y . Jin, “A strengthened dominance relation considering convergence and diversity for evolution- ary many-objective optimization,”IEEE Transactions on Evolutionary Computation, vol. 23, no. 2, pp. 331–345, 2019

    2019

  9. [17]

    Utopian point based decomposition for multi-objective optimization problems with complicated Pareto fronts,

    Y . Qi, Q. Zhang, X. Ma, Y . Quan, and Q. Miao, “Utopian point based decomposition for multi-objective optimization problems with complicated Pareto fronts,”Applied Soft Computing, vol. 61, pp. 844– 859, 2017

    2017

  10. [18]

    A generator for multiobjective test problems with difficult-to-approximate Pareto front boundaries,

    Z. Wang, Y .-S. Ong, J. Sun, A. Gupta, and Q. Zhang, “A generator for multiobjective test problems with difficult-to-approximate Pareto front boundaries,”IEEE Transactions on Evolutionary Computation, vol. 23, no. 4, pp. 556–571, 2019

    2019

  11. [19]

    Reference point specification in MOEA/D for multi-objective and many-objective problems,

    H. Ishibuchi, K. Doi, and Y . Nojima, “Reference point specification in MOEA/D for multi-objective and many-objective problems,” inIEEE International Conference on Systems, Man, and Cybernetics (SMC). IEEE, 2016, pp. 004 015–004 020

    2016

  12. [20]

    On the effect of reference point in MOEA/D for multi-objective optimization,

    R. Wang, J. Xiong, H. Ishibuchi, G. Wu, and T. Zhang, “On the effect of reference point in MOEA/D for multi-objective optimization,”Applied Soft Computing, vol. 58, pp. 25–34, 2017

    2017

  13. [21]

    Decomposition-based multi-objective evolutionary algorithm with model-based ideal point estimation,

    Y . Wu, R. Zheng, and Z. Wang, “Decomposition-based multi-objective evolutionary algorithm with model-based ideal point estimation,” in Proceedings of the Genetic and Evolutionary Computation Conference (GECCO), 2023, pp. 768–776

    2023

  14. [22]

    An easy-to-use real-world multi-objective optimization problem suite,

    R. Tanabe and H. Ishibuchi, “An easy-to-use real-world multi-objective optimization problem suite,”Applied Soft Computing, vol. 89, p. 106078, 2020

    2020

  15. [23]

    K. Deb, L. Thiele, M. Laumanns, and E. Zitzler,Scalable test problems for evolutionary multiobjective optimization. Springer, 2005

    2005

  16. [24]

    Generalized center method for multiobjective engineering optimization,

    F. Cheng and X. Li, “Generalized center method for multiobjective engineering optimization,”Engineering Optimization, vol. 31, no. 5, pp. 641–661, 1999

    1999

  17. [25]

    Multiobjective evolutionary algorithms: A comparative case study and the strength Pareto approach,

    E. Zitzler and L. Thiele, “Multiobjective evolutionary algorithms: A comparative case study and the strength Pareto approach,”IEEE Trans- actions on Evolutionary Computation, vol. 3, no. 4, pp. 257–271, 1999

    1999

  18. [26]

    Biased multiobjective optimization and decomposition algorithm,

    H. Li, Q. Zhang, and J. Deng, “Biased multiobjective optimization and decomposition algorithm,”IEEE Transactions on Cybernetics, vol. 47, no. 1, pp. 52–66, 2017

    2017

  19. [27]

    Multi-objective genetic algorithms: Problem difficulties and construction of test problems,

    K. Deb, “Multi-objective genetic algorithms: Problem difficulties and construction of test problems,”Evolutionary Computation, vol. 7, no. 3, pp. 205–230, 1999

    1999

  20. [28]

    A review of multiobjective test problems and a scalable test problem toolkit,

    S. Huband, P. Hingston, L. Barone, and L. While, “A review of multiobjective test problems and a scalable test problem toolkit,”IEEE Transactions on Evolutionary Computation, vol. 10, no. 5, pp. 477–506, 2006

    2006

  21. [29]

    Multi- objective optimization of cellular fenestration by an evolutionary algo- rithm,

    J. A. Wright, A. Brownlee, M. M. Mourshed, and M. Wang, “Multi- objective optimization of cellular fenestration by an evolutionary algo- rithm,”Journal of Building Performance Simulation, vol. 7, no. 1, pp. 33–51, 2014

    2014

  22. [30]

    Is MO-CMA-ES superior to NSGA- II for the evolution of multi-objective neuro-controllers?

    A. Moshaiov and O. Abramovich, “Is MO-CMA-ES superior to NSGA- II for the evolution of multi-objective neuro-controllers?” inIEEE Congress on Evolutionary Computation (CEC). IEEE, 2014, pp. 2809– 2816

    2014

  23. [31]

    Decomposition of a multiobjective optimization problem into a number of simple multiobjective subprob- lems,

    H.-L. Liu, F. Gu, and Q. Zhang, “Decomposition of a multiobjective optimization problem into a number of simple multiobjective subprob- lems,”IEEE Transactions on Evolutionary Computation, vol. 18, no. 3, pp. 450–455, 2014

    2014

  24. [32]

    Challenging test problems for multi-and many-objective optimization,

    S. Zapotecas-Mart ´ınez, C. A. C. Coello, H. E. Aguirre, and K. Tanaka, “Challenging test problems for multi-and many-objective optimization,” Swarm and Evolutionary Computation, vol. 81, p. 101350, 2023

    2023

  25. [33]

    A mul- tiobjective test suite with hexagon Pareto fronts and various feasible regions,

    T. Matsumoto, N. Masuyama, Y . Nojima, and H. Ishibuchi, “A mul- tiobjective test suite with hexagon Pareto fronts and various feasible regions,” inIEEE Congress on Evolutionary Computation (CEC). IEEE, 2019, pp. 2058–2065

    2019

  26. [34]

    Investigating the effect of imbalance between convergence and diversity in evolutionary multiobjective algorithms,

    H.-L. Liu, L. Chen, K. Deb, and E. D. Goodman, “Investigating the effect of imbalance between convergence and diversity in evolutionary multiobjective algorithms,”IEEE Transactions on Evolutionary Compu- tation, vol. 21, no. 3, pp. 408–425, 2017

    2017

  27. [35]

    A set of new multi-and many- objective test problems for continuous optimization and a comprehensive experimental evaluation,

    Y . Zhou, X. He, Y . Xiang, and S. Cai, “A set of new multi-and many- objective test problems for continuous optimization and a comprehensive experimental evaluation,”Artificial Intelligence, vol. 276, pp. 105–129, 2019

    2019

  28. [36]

    Solving difficult constrained optimization problems by theεconstrained differential evolution with gradient-based mutation,

    T. Takahama and S. Sakai, “Solving difficult constrained optimization problems by theεconstrained differential evolution with gradient-based mutation,”Constraint-Handling in Evolutionary Optimization, pp. 51– 72, 2009

    2009

  29. [37]

    An analysis on recombination in multi- objective evolutionary optimization,

    C. Qian, Y . Yu, and Z.-H. Zhou, “An analysis on recombination in multi- objective evolutionary optimization,”Artificial Intelligence, vol. 204, pp. 99–119, 2013

    2013

  30. [38]

    Crossover can guarantee expo- nential speed-ups in evolutionary multi-objective optimisation,

    D.-C. Dang, A. Opris, and D. Sudholt, “Crossover can guarantee expo- nential speed-ups in evolutionary multi-objective optimisation,”Artificial Intelligence, p. 104098, 2024

    2024

  31. [39]

    Failure of Pareto-based moeas: Does non-dominated really mean near to optimal?

    K. Ikeda, H. Kita, and S. Kobayashi, “Failure of Pareto-based moeas: Does non-dominated really mean near to optimal?” inIEEE Congress on Evolutionary Computation (CEC), vol. 2. IEEE, 2001, pp. 957–962

    2001

  32. [40]

    On scalable multiobjective test problems with hardly dominated boundaries,

    Z. Wang, Y .-S. Ong, and H. Ishibuchi, “On scalable multiobjective test problems with hardly dominated boundaries,”IEEE Transactions on Evolutionary Computation, vol. 23, no. 2, pp. 217–231, 2019

    2019

  33. [41]

    Solving many-objective optimization problems by a Pareto-based evolutionary algorithm with preprocessing and a penalty mechanism,

    Y . Liu, N. Zhu, and M. Li, “Solving many-objective optimization problems by a Pareto-based evolutionary algorithm with preprocessing and a penalty mechanism,”IEEE Transactions on Cybernetics, vol. 51, no. 11, pp. 5585–5594, 2021

    2021

  34. [42]

    Miettinen,Nonlinear multiobjective optimization

    K. Miettinen,Nonlinear multiobjective optimization. Springer, 1998

    1998

  35. [43]

    PSA-CMA-ES: CMA-ES with popula- tion size adaptation,

    K. Nishida and Y . Akimoto, “PSA-CMA-ES: CMA-ES with popula- tion size adaptation,” inProceedings of the Genetic and Evolutionary Computation Conference (GECCO), 2018, pp. 865–872

    2018

  36. [44]

    The CMA evolution strategy: A tutorial,

    N. Hansen, “The CMA evolution strategy: A tutorial,”arXiv preprint arXiv:1604.00772, 2016

    2016 arXiv

  37. [45]

    Benchmarking the PSA-CMA-ES on the BBOB noiseless testbed,

    K. Nishida and Y . Akimoto, “Benchmarking the PSA-CMA-ES on the BBOB noiseless testbed,” inProceedings of the Genetic and Evolution- ary Computation Conference (GECCO), 2018, pp. 1529–1536

    2018

  38. [46]

    Warm starting CMA-ES for hyperparameter optimization,

    M. Nomura, S. Watanabe, Y . Akimoto, Y . Ozaki, and M. Onishi, “Warm starting CMA-ES for hyperparameter optimization,” inProceedings of the AAAI conference on artificial intelligence (AAAI), vol. 35, no. 10, 2021, pp. 9188–9196. 15

    2021

  39. [47]

    Injecting external solutions into CMA-ES,

    N. Hansen, “Injecting external solutions into CMA-ES,”arXiv preprint arXiv:1110.4181, 2011

    2011 arXiv

  40. [48]

    Injecting CMA-ES into MOEA/D,

    S. Zapotecas-Mart ´ınez, B. Derbel, A. Liefooghe, D. Brockhoff, H. E. Aguirre, and K. Tanaka, “Injecting CMA-ES into MOEA/D,” inPro- ceedings of the Genetic and Evolutionary Computation Conference (GECCO), 2015, pp. 783–790

    2015

  41. [49]

    Using well- understood single-objective functions in multiobjective black-box op- timization test suites,

    D. Brockhoff, A. Auger, N. Hansen, and T. Tu ˇsar, “Using well- understood single-objective functions in multiobjective black-box op- timization test suites,”Evolutionary Computation, vol. 30, no. 2, pp. 165–193, 2022

    2022

  42. [50]

    PlatEMO: A MATLAB plat- form for evolutionary multi-objective optimization [educational forum],

    Y . Tian, R. Cheng, X. Zhang, and Y . Jin, “PlatEMO: A MATLAB plat- form for evolutionary multi-objective optimization [educational forum],” IEEE Computational Intelligence Magazine, vol. 12, no. 4, pp. 73–87, 2017

    2017

  43. [51]

    A generalized scalarization method for evo- lutionary multi-objective optimization,

    R. Zheng and Z. Wang, “A generalized scalarization method for evo- lutionary multi-objective optimization,” inProceedings of the AAAI Conference on Artificial Intelligence (AAAI), vol. 37, no. 10, 2023, pp. 12 518–12 525

    2023

  44. [52]

    Adaptive re- placement strategies for MOEA/D,

    Z. Wang, Q. Zhang, A. Zhou, M. Gong, and L. Jiao, “Adaptive re- placement strategies for MOEA/D,”IEEE Transactions on Cybernetics, vol. 46, no. 2, pp. 474–486, 2016

    2016

  45. [53]

    Hypervolume- based cooperative coevolution with two reference points for multi- objective optimization,

    L. M. Pang, H. Ishibuchi, L. He, K. Shang, and L. Chen, “Hypervolume- based cooperative coevolution with two reference points for multi- objective optimization,”IEEE Transactions on Evolutionary Computa- tion, vol. 28, no. 4, pp. 21 054 – 1068, 2024

    2024

  46. [54]

    SMS-EMOA: Multiobjec- tive selection based on dominated hypervolume,

    N. Beume, B. Naujoks, and M. Emmerich, “SMS-EMOA: Multiobjec- tive selection based on dominated hypervolume,”European Journal of Operational Research, vol. 181, no. 3, pp. 1653–1669, 2007

    2007

  47. [55]

    Multiobjective optimization problems with com- plicated Pareto sets, MOEA/D and NSGA-II,

    H. Li and Q. Zhang, “Multiobjective optimization problems with com- plicated Pareto sets, MOEA/D and NSGA-II,”IEEE Transactions on Evolutionary Computation, vol. 13, no. 2, pp. 284–302, 2009

    2009

  48. [56]

    A survey of multi-objective evolutionary algorithm based on decomposition: Past and future,

    K. Li, “A survey of multi-objective evolutionary algorithm based on decomposition: Past and future,”IEEE Transactions on Evolutionary Computation, 2024, in press

    2024

  49. [57]

    Performance of decomposition-based many-objective algorithms strongly depends on Pareto front shapes,

    H. Ishibuchi, Y . Setoguchi, H. Masuda, and Y . Nojima, “Performance of decomposition-based many-objective algorithms strongly depends on Pareto front shapes,”IEEE Transactions on Evolutionary Computation, vol. 21, no. 2, pp. 169–190, 2017

    2017

  50. [58]

    What weights work for you? Adapting weights for any Pareto front shape in decomposition-based evolutionary multi- objective optimisation,

    M. Li and X. Yao, “What weights work for you? Adapting weights for any Pareto front shape in decomposition-based evolutionary multi- objective optimisation,”Evolutionary Computation, vol. 28, no. 2, pp. 227–253, 2020. 16 This is supplementary material for “Enhanced Ideal Ob- j...

    2020

Pith tools

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