Deep Multiple Clustering: From Foundations to Context-Aware Approaches
TL;DR
This tutorial explains why real-world datasets often inherently possess multiple latent structures and demonstrate how deep multiple clustering methods uncover these alternative perspectives, and critically discuss evaluation metrics for clustering quality and diversity.
Abstract
Deep Multiple clustering generalizes the unsupervised deep learning paradigm from identifying a single universal partition of data to discovering several distinct yet meaningful groupings. This tutorial provides a comprehensive overview of deep multiple clustering techniques, with an emphasis on theoretical foundations, methodological comparisons, and applications. We explain why real-world datasets often inherently possess multiple latent structures and demonstrate how deep multiple clustering methods uncover these alternative perspectives. Throughout the discussion, from fundamental deep multiple clustering methods to recent advances in context-aware approaches, we also highlight real-world applications in vision, text, and biology, critically discuss evaluation metrics for clustering quality and diversity, and identify open theoretical and practical challenges. The tutorial is self-contained and interactive, suitable for a broad KDD audience interested in unsupervised learning, representation learning, and context-aware data exploration.