.png){kind=icon}
Return
HVC-RankNet: A Hypervolume Contribution Ranking Network Embedded in Reference-Vector-Based Algorithms for Many-Objective Optimization Problems
.png){kind=icon}
PRE
AI
DOI：10.1109/tevc.2026.3724941.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Reference-vector-based many-objective evolutionary algorithms (MaOEAs) often lack an explicit mechanism to evaluate the marginal contributions of candidate solutions relative to the current selected solution set within each cluster. This limitation may negatively affect the coverage quality of the Pareto front approximation. To this end, the intra-cluster competition is formulated as a set-conditioned marginal-gain ranking problem, and a generic hypervolume contribution (HVC)-guided intra-cluster selection operator is proposed for reference-vector-based MaOEAs. To further improve the efficiency of this operator, an HVC Ranking Network (HVC-RankNet) is developed as an asymmetric context-aware ranking model tailored to this formulation for batch-wise ranking of intra-cluster candidate solutions. In addition, theoretical analyses are provided to justify the proposed operator and ranking model. Experimental results show that HVC-RankNet achieves leading or competitive ranking accuracy in most evaluated settings. When embedded in five reference-vector-based MaOEAs, the proposed operator mainly improves HV-oriented coverage and often yields favorable IGD+ results, whereas its effects on uniformity-oriented diversity and runtime are less consistent.
Keywords:
Hypervolume contribution
intra-cluster selection
learning-to-rank
many-objective optimization
reference-vector-based algorithms
Journal
IF:
12
Papers:
1.9K
Citations:
2.4W
Organization
Cited Papers
No cited papers available