Proposal Description

The proposed session “Advances in Soft and Fuzzy Clustering” brings together recent methodological advances that extend the scope, flexibility, and interpretability of clustering for modern complex data. Unlike hard clustering methods, which assign each observation to a single group, soft and fuzzy clustering methods allow observations to have partial membership in multiple groups. This feature is particularly valuable in real-world applications, where clusters may overlap, and observations may display characteristics of more than one sub-population.
The session highlights three complementary directions in current clustering research. The first focuses on L0-regularized fuzzy clustering, which promotes sparsity in the membership matrix and provides a principled bridge between hard and soft cluster assignments. The second develops a matrix-variate Mahalanobis-type fuzzy clustering approach, extending adaptive distance-based fuzzy clustering to matrix-valued observations while preserving their natural row-column structure. The third contribution introduces parsimonious dimension-wise scaled normal mixtures for robust model-based clustering, addressing data in which different variables may exhibit different degrees of tail heaviness.
Together, these talks show how soft and fuzzy clustering can be adapted to heterogeneous, high-dimensional, and structured datasets while maintaining robustness, computational practicality, and interpretability. The session will be of interest to researchers working on clustering methodology, model-based classification, fuzzy methods, and applications involving complex data structures.