An optimization of the dynamic set of clusters for similarity searches

• Autores: Samat, Pablo; Ludueña, Verónica; Reyes, Nora Susana; Paredes, Rodrigo; Figueroa, Karina; Lagos, Miguel
• Año de publicación: 2025
• Idioma: inglés
• Tipo de recurso: documento de conferencia
• Materia: Ciencias Informáticas; similarity searches; dynamic indexes; non conventional databases
• Institución: Universidad Nacional de La Plata

While metric space searching offers a promising solution for efficient similarity searches, the approach lags behind traditional databases in several key areas. A primary weakness is that most indexing methods are static; very few can handle dynamic insertions and deletions cost-effectively without degrading performance. Additionally, even fewer schemes are optimized to operate efficiently on secondary storage. The Dynamic Set of Clusters (DSC) is a dynamic index built for secondary memory, but its performance is hindered by a lack of cluster compactness. To overcome this, we propose an optimization that uses "cut regions” established by global pivots, to more accurately define the data-containing zones within each cluster. This enhanced method is competitive with state-of-the-art approaches, performs well in secondary memory, and serves as a practical alternative for numerous database applications.
Red de Universidades con Carreras en Informática