Metaheurísticas inspiradas na natureza aceleradas por gpu para agrupamento de dados com múltiplas visões

Abstract

Multi-view clustering has emerged as a powerful framework for integrating heterogeneous feature spaces into a unified learning process. Nevertheless, the optimization of multi-view models remains computationally demanding, particularly when employing population-based metaheuristics on large datasets. This paper presents GPU-accelerated adaptations of six metaheuristic algorithms — Particle Swarm Optimization (PSO), Cuckoo Search (CS), Firefly Algorithm (FA), Elephant Herding Optimization (EHO), Bees Algorithm (BA), and Black Hole (BH) — applied to multi-view clustering. Each algorithm was reformulated to operate on a unified centroid representation and optimized under two objective functions: a full version derived from TW-K-Means and a simplified view-weighted version (VWOF). The proposed implementations exploit GPU parallelism and vectorized computation to achieve substantial reductions in runtime while maintaining clustering quality. Experiments on seven benchmark datasets demonstrate consistent improvements over traditional multi-view clustering methods in both clustering accuracy (ARI, NMI, Purity) and computational efficiency, achieving speedups of up to 50×. Statistical analyses confirm the significance of the observed differences, reinforcing the potential of GPU acceleration for metaheuristic-based multi-view clustering.