Evolving Clusters: A Dynamical Perspective on Time-Dependent Mixture Models
Please login to view abstract download link
We propose a novel framework for evolutionary clustering based on mean field theory. Traditional soft clustering methods, such as finite mixture models with Expectation-Maximization (EM), struggle in dynamic settings where data evolves over time. Our approach models time-dependent mixture densities via a nonlinear Fokker-Planck equation, coupling EM updates with a dynamical system to ensure temporal stability. This yields clusterings that remain statistically accurate while capturing long-term trends, offering a robust and interpretable solution for dynamic data analysis.
