A Bayesian Approach to the Changepoint Problem in Signal Processing

J. Portela (Spain)


Changepoint problem, Bayesian models, signal processing, exponentialpower distribution.


In this work the problem of detecting changepoints in a sig nal row of data is adressed. This is a common problem arising in signal processing and other areas as Economics or Medicine. A fully Bayesian perspective is adopted, and the signal noise modeling is realized through a symmetric generalization of Gaussian and Laplace approaches used in previous research, revealing that varying the model of noise in different sections of the data can be more accu rate than simpler models used in previous research. Due to the kind of computational work needed in this model, numerical methods are applied instead of the usual Gibbs sampling approach to the estimation of the changepoint lo cation. The multiple changepoint problem is also targeted in our research, and an application to well-log data has been presented, conformingly.

Important Links:

Go Back