Riemannian Mean of Positive Definite Matrices

Channel:
Microsoft Research
Subscribers:
349,000
Published on ● Video Link: https://www.youtube.com/watch?v=S26Kz1DuV5w


Duration: 54:57
587 views
12

Positive definite matrices arise in several contexts: quantum mechanics, statistics, machine learning, image processing, elasticity etc. Various notions of averaging are useful in different contexts. In the 1970's physicists, electrical engineers, and matrix theorists developed a notion of a geometric mean of two positive definite matrices. The problem of extending the concept to more than two matrices remained open for several years. The problem was resolved in 2004 using ideas from Riemannian geometry. Meanwhile, effective use of the idea has been made in applications in radar data processing and imaging of the brain. In this talk, I will explain the main ideas from the perspective of matrix analysis. Reference: R. Bhatia, Positive Definite Matrices, Princeton University Press 2007, Hindustan Book Agency 2007.



Other Videos By Microsoft Research


2016-08-11Scalable Inference of Attributes in Entity-Relationship Graphs
2016-08-11Making SVMs Robust to Uncertainty in Kernel Matrices
2016-08-11Weakness Can Be Quarantined!
2016-08-11Noisy, Missing and Corrupted Data
2016-08-11Semi-Supervised Structured Output Learning
2016-08-11Functional Programming in the Wild
2016-08-11Majorana Modes, Non-Abelian Anyons, and Topological Quantum Computation
2016-08-11Clustering
2016-08-11Lines, Shading, and the Perception of 3D Shape
2016-08-11StreamSVM: Linear SVMs When Data Does Not Fit In Memory
2016-08-11Riemannian Mean of Positive Definite Matrices
2016-08-11A Proximal Point Algorithm For Log-Determinant Optimization With Group Lazzo Regularization
2016-08-11Functional Site Prediction by Exploiting Correlations between Labels of Interacting Residues
2016-08-11Predicting (and Understanding) the 2012 Election
2016-08-11Small Variance Asymptotics, Bayesian Nonparametrics, and k-means
2016-08-11Probabilistic Model for Scene Text Recognition
2016-08-11The Changing Landscape of Parallel Computing ΓÇô II
2016-08-11Computing Elliptic Curves over Q(√5)
2016-08-11The Changing Landscape of Parallel Computing - Tools (Testing and Debugging)
2016-08-11ChronoZoom: Bridging the Gap Between the Humanities and Sciences
2016-08-11Robust Subspace Modeling


Tags:
microsoft research