Otimizando o Cálculo do Índice de Silhueta através de Paralelismo em GPU

• Thiago B. Lopes UFG
• Wellington S. Martins UFG

DOI: https://doi.org/10.5753/eradco.2023.233879

Abstract

Meta-features are fundamental for optimizing machine learning algorithms, but the calculation of meta-features related to the clustering set is computationally costly due to the complex analyses of clustering properties. In this paper, we present an approach that employs GPU parallelism to increase the efficiency of calculating the silhouette index, a key metric in clustering, making it more viable for large datasets.

References

Brazdil, P., Giraud-Carrier, C., Soares, C., and Vilalta, R. (2009). Metalearning: Applications to data mining. Springer Publishing Company.

Google (2021). Google colaboratory. [link]. Accessed: 2023-05-25.

Luna-Romera, J. M., Martínez Ballesteros, M., García-Gutiérrez, J., and Riquelme, J. (2016). An approach to silhouette and dunn clustering indices applied to big data in spark. Advances in Artificial Intelligence. CAEPIA 2016. Lecture Notes in Computer Science(), vol 9868. Springer, Cham.

Rousseeuw, P. (1987). Silhouettes: a graphical aid to the interpretation and validation of cluster analysis. Journal of Computational and Applied Mathematics 20, 53-65.

Silva, L. L., Franco, R., Carvalho, A., and Martins, W. (2023). Gpu acceleration of clustering meta-feature extractionusing rapids. XXII Workshop em Desempenho de Sistemas Computacionais e de Comunicacão, (wperformance 2023) edition.