Unsupervised Classification

Unsupervised classification aims to group unlabeled data into meaningful categories without relying on pre-defined classes, a crucial task in numerous fields facing data scarcity or high annotation costs. Current research emphasizes leveraging powerful deep learning architectures, including large visual and language models, spiking neural networks, and transformer-based models, often incorporating techniques like contrastive learning and prototype-based methods to improve clustering accuracy and robustness. These advancements are significantly impacting diverse applications, from analyzing social media images and medical claims to classifying satellite imagery for agricultural monitoring and identifying patterns in complex datasets like those from space physics simulations. The development of improved internal validation indices is also a key area of focus to ensure the selection of optimal algorithms for specific datasets.