A Hidden Markov Filtering Approach to Multiple Change-point Models

We describe a hidden Markov modeling approach to multiple change-points that has attractive computational and statistical properties. This approach yields explicit recursive filters and smoothers for estimating the piecewise constant parameters. Applications to array-CGH data analysis in genetic studies of cancer and to on-line detection, estimation and adaptive control of stochastic systems whose parameters may undergo occasional changes are given to illustrate the versatility of the proposed methodology.