Scalable Bayesian Multiple Changepoint Detection via Auxiliary Uniformization

TL;DR

This paper introduces a scalable Bayesian method for multiple changepoint detection in discrete data by transforming it into a continuous-time problem using auxiliary event times, enabling efficient inference even with long sequences.

Contribution

It proposes a novel formulation with auxiliary event times and a quadratic-scaling FFBS algorithm, allowing efficient Bayesian changepoint detection with unbounded numbers of changepoints.

Findings

Quadratic scaling reduces computational costs for long sequences.

Method effectively detects multiple changepoints in simulations.

Real data analysis confirms practical applicability.

Abstract

By attaching auxiliary event times to the chronologically ordered observations, we formulate the Bayesian multiple changepoint problem of discrete-time observations into that of continuous-time ones. A version of forward-filtering backward-sampling (FFBS) algorithm is proposed for the simulation of changepoints within a collapsed Gibbs sampling scheme. Ideally, both the computational cost and memory cost of the FFBS algorithm can be quadratically scaled down to the number of changepoints, instead of the number of observations, which is otherwise prohibitive for a long sequence of observations. The new formulation allows the number of changepoints accrue unboundedly upon the arrivals of new data. Also, a time-varying changepoint recurrence rate across different segments is assumed to characterize diverse scales of run lengths of changepoints. We then suggest a continuous-time Viterbi algorithm for obtaining the Maximum A Posteriori (MAP) estimates of changepoints. We demonstrate the methods through simulation studies and real data analysis.