The Derivative of the Matrix Geometric Mean with an Application to the Nonnegative Decomposition of Tensor Grids
We provide an expression for the derivative of the weighted matrix geometric mean, with respect to both the matrix arguments and the weights, that can be easily translated to an algorithm for its computation. As an application, we consider the problem of the approximate decomposition of a tensor grid M, a matrix whose entries are positive definite matrices. For different geometries on the set of positive definite matrices, we derive an approximate decomposition such that any column of M is a barycentric combination of the columns of a smaller tensor grid. This extends the Euclidean case, already considered in the literature, to the geometry in which the barycenter is the matrix geometric mean and the log-Euclidean geometry.
KeywordsMatrix geometric mean Karcher mean Tensor grid Positive definite matrix Nonnegative factorization
The authors would like to thank the referees for carefully reading the manuscript, providing many insightful comments which improved the presentation of the chapter.
- 10.Higham, N.J.: Functions of Matrices: Theory and Computation. Society for Industrial and Applied Mathematics (SIAM), Philadelphia (2008)Google Scholar
- 17.Lee, D.D., Seung, H.S.: Algorithms for non-negative matrix factorization. In: Leen, T., Dietterich, T., Tresp, V. (eds) Advances in Neural Information Processing Systems, vol. 13, pp. 556–562. MIT Press, Cambridge (2001)Google Scholar
- 21.Rentmeesters, Q., Absil, P.A.: Algorithm comparison for Karcher mean computation of rotation matrices and diffusion tensors. In: 2011 19th European Signal Processing Conference, pp. 2229–2233. IEEE (2011)Google Scholar
- 22.Sra, S., Dhillon, I.S.: Nonnegative Matrix Approximation: Algorithms and Applications. Technical Report TR-06-27, The University of Texas at Austin, June (2006)Google Scholar
- 23.Wang, Y., Salehian, H., Cheng, G., Vemuri, B.C.: Tracking on the product manifold of shape and orientation for tractography from diffusion MRI. In: 2014 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 3051–3056. IEEE (2014)Google Scholar
- 24.Xie, Y., Ho, J., Vemuri, B.: Nonnegative factorization of diffusion tensor images and its applications. In: International Conference on Information Processing in Medical Imaging (IPMI) (2011)Google Scholar