Fast Variational Inference for Dirichlet Process Mixture Models

Nonparametric Bayesian Models currently suffer from a lack of efficient infer-ence algorithms. This precludes the applicability of these methods when real-time analysis is needed. In this work we present a fast on-line variational inference al-gorithm for Dirichlet Process Mixture Models which takes advantage of the evo-lution of the data to speed up inference. The method has been applied to perform dynamic clustering on neuronal activity data in order to identify clusters having similar behaviour.