Subsampling-based HMC parameter estimation with application to large datasets classification

Abstract

Your article is protected by copyright and all rights are held exclusively by Springer-Verlag London Limited. This e-offprint is for personal use only and shall not be self-archived in electronic repositories. If you wish to self-archive your article, please use the accepted manuscript version for posting on your own website. You may further deposit the accepted manuscript version in any repository, provided it is only made publicly available 12 months after official publication or later and provided acknowledgement is given to the original source of publication and a link is inserted to the published article on Springer's website. The link must be accompanied by the following text: "The final publication is available at link.springer.com". Abstract This paper presents a contextual algorithm for the approximation of Baum's forward and backward probabilities , which are extensively used in the framework of Hidden Markov chain models for parameter estimation. The method differs from the original algorithm by taking into account only a neighborhood of limited length and not all the data in the chain for computations. It then becomes possible to propose a bootstrap subsampling strategy for the computation of forward and backward probabilities, which greatly reduces computation time and memory saving required for EM-based parameter estimation. Comparative experiments regarding the neighborhood size and the bootstrap sample size are conducted by mean of unsupervised classification error rates. Practical interest of such an algorithm is then illustrated through the segmentation of large-size images; classification results confirm the validity and the accuracy of the proposed algorithm while greatly reducing computation and memory requirements.