Computational Information Geometry Wonderland

The previous post was about computing the centroid of normals with respect to the symmetrized Kullback-Leibler divergence. Today, I would like to present the same problem for a different measure: The Bhattacharyya distortion measure. Given two normals in d dim, their distance is expressed by:

The authors of

http://www.citeulike.org/user/ciga/article/1823637 An optimal Bhattacharyya centroid algorithm for Gaussian clusteringwith applications in automatic speech recognition

present an iterative algorithm. This is the same flavor as for the SKL case.