Cohort Organized Learning: Clustering Through Agreement

Cohort Organized Learning (CoOL) is a novel clustering method that operates without explicit distance or similarity computations, utilizing neural networks to estimate clusters instead. The paper details the derivation of gradients via expectation maximization for training, convergence monitoring techniques, and evaluation of clusters post-training, with applications demonstrated on vector data and images. This approach offers a flexible clustering solution for practitioners by enabling the handling of diverse data types while addressing potential limitations and future applications.