Optimisation de Fuzzy C-Means (FCM) clustering par la méthode des directions alternées (ADMM) - Archive ouverte HAL
Communication Dans Un Congrès Année : 2023

Optimization of Fuzzy C-Means (FCM) clustering with the Alternating Direction Method of Multiplier (ADMM)

Optimisation de Fuzzy C-Means (FCM) clustering par la méthode des directions alternées (ADMM)

Résumé

Among the clustering methods, K-Means and variants are very popular. These methods solve at each iteration the first order optimality conditions. However, in some cases, the function to be minimized is not convex, as for the Fuzzy C-Means version with Mahalanobis distance (FCM-GK). In this study, we apply the Alternating Directions Method of Multiplier (ADMM) to ensure a good convergence. ADMM is often applied to solve a separable convex minimization problem with linear constraints. ADMM is a decomposition/coordination method with a coordination step provided by Lagrange multipliers. By appropriately introducing auxiliary variables, this method allows the problem to be decomposed into easily solvable convex subproblems while keeping the same iterative structure. Numerical results have demonstrated the significant perfor- mance of the proposed method compared to the standard method especially for high dimensional data.
Fichier principal
Vignette du fichier
Optimisation_de_Fuzzy_C_Means__FCM__clustering_par_la_m_thode_des_directions_altern_es__ADMM____EGC-2.pdf (112.62 Ko) Télécharger le fichier

Dates et versions

hal-03974754 , version 1 (24-02-2023)

Identifiants

  • HAL Id : hal-03974754 , version 1

Citer

Benoit Albert, Violaine Antoine, Jonas Koko. Optimisation de Fuzzy C-Means (FCM) clustering par la méthode des directions alternées (ADMM). Extraction et Gestion des Connaissances (EGC), 2023, Lyon, France. pp.247-258. ⟨hal-03974754⟩
61 Consultations
100 Téléchargements

Partager

More