Reviews: A Statistical Recurrent Model on the Manifold of Symmetric Positive Definite Matrices

The submission #5332, entitled "Statistical Recurrent Models on Manifold valued Data", presents a framework for the fit of Recurrent Neural networks on SPD matrices. Much of the efforts are spent on deriving the framework for these computations. Particular attention is paid to the estimation of Fréchet mean for this type of data, in order to obtain a fast yet reliable estimators. All these data are very low-dimensional, so that the reported algorithmic optimization unlikely to be useful in practical settings. The experiments show that the proposed approach is accurate in the problems considered and requires fewer iterations for convergence.