Regresar

Investigating the Climbing Mechanism of the Weight-Guided Random Bit Climber for Many-Objective Optimization

Abstract:

In this paper, we investigate a Weight-Guided Random Bit Climbing algorithm (wgRBC) for tackling multi and many-objective binary epistactic optimization problems. The algorithm integrates scalarizing functions with a random bit climber and leverages predefined weights to explore diverse search directions while maintaining a well-distributed reference population. By performing weight-guided hill climbing, wgRBC combines global exploration in the direction of the weights with local search through hill climbing. This fusion of global and local search enables the algorithm to explore widely distributed yet uniformly spaced solutions in objective space. The hill climbing process is analyzed, including the roles of weights in reaching local optimum. To evaluate the effectiveness of wgRBC, extensive experiments were conducted on MNK-landscapes with varying levels of epistasis and objectives, comparing wgRBC against established methods such as dRBC, NSGA-III, and MOEA/D. The results show the superiority of wgRBC in achieving higher hypervolume (HV) across almost all objectives, particularly for problems with moderate to high epistasis (K > 1). Although wgRBC is simpler than decomposition-based algorithms, it consistently outperforms them in most scenarios, highlighting its potential for scalable optimization in challenging binary landscapes.