Multinomial Sampling for Hierarchical Change-Point Detection

TL;DR

This paper introduces a multinomial sampling method to enhance hierarchical change-point detection in high-dimensional time-series, improving accuracy and reducing delay while maintaining computational efficiency.

Contribution

It proposes a novel multinomial sampling approach that improves change-point detection in latent variable models, addressing uncertainty issues in pseudo-observations.

Findings

Outperforms baseline methods in detection accuracy

Reduces detection delay in experiments

Demonstrates effectiveness in a human behavior study

Abstract

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