WASHH: Reinventing Black-Box Optimization with Whales and Heuristics

Learning-assisted algorithm design faces a fundamental challenge: making reliable decisions with minimal computational resources. Enter WASHH, a Whale-guided Adaptive Selection Hyper-Heuristic that aims to revolutionize how we tackle continuous black-box optimization. Combining the strengths of different metaheuristic techniques, WASHH is designed to maximize performance while minimizing evaluations.

what's WASHH?

At its core, WASHH employs the Whale Optimization Algorithm (WOA) as its primary exploitation mechanism. However, it doesn't stop there. It integrates a variety of strategies like PSO-style memory, GWO-style leader averaging, DE-style variation, local coordinate search, and anchor-guided refinement. This diverse toolkit allows WASHH to adaptively select the most promising search behavior based on current needs.

Performance and Evaluation

WASHH was put to the test on a suite of ten 30-dimensional benchmark functions, running 10 independent trials with a budget of 12,000 evaluations. The results? Impressive. WASHH achieved an average rank of 1.10, either outperforming or matching the best results across all benchmarks. It improved over the standard WOA on eight functions and tied at the numerical optimum on the Rastrigin and Griewank functions. The paper's key contribution: demonstrating that a hyper-heuristic approach can effectively use anchor guidance to refine decision-making processes in LEAD systems.

Real-World Implications

Why does this matter? Consider the practical application tested: configuring hyperparameters for breast cancer diagnosis with a mere 300-evaluation budget. Here, WASHH delivered the lowest mean validation log loss among its peers. That suggests its potential as a lightweight, efficient tool for real-world problems where computational resources are limited.

Is WASHH the Future?

Could WASHH chart the course for future advancements in learning-assisted algorithm design? It certainly raises the question of whether combining multiple heuristic strategies can become the norm. The adaptability and efficiency demonstrated by WASHH could set a new standard, especially for applications constrained by evaluation budgets.

The ablation study reveals the individual contributions of each metaheuristic component within WASHH, underscoring its sophisticated design. Yet, while the results are promising, further exploration in diverse domains is needed to solidify its place in the optimization toolkit.