Multinomial Sampling of Latent Variables for Hierarchical Change-Point Detection

Abstract Bayesian change-point detection, with latent variable models, allows to perform segmentation of high-dimensional time-series heterogeneous statistical nature. We assume that change-points lie on a lower-dimensional manifold where we aim infer discrete representation via subsets variables. For this particular model, full inference is computationally unfeasible and pseudo-observations based point-estimates variables are used instead. However, if their estimation not certain enough, detection gets affected. To circumvent problem, propose multinomial sampling methodology improves the rate reduces delay while keeping complexity stable analytically tractable. Our experiments show results outperform baseline method also provide an example oriented human behavioral study.