* Title: Matrix Manifold Optimization for Gaussian Mixtures
* Speaker: TaeHyun Hwang (SKKU)
* Place/Date: 32356A / 13 October 2017 (Friday) am 10:00-12:00
* Abstract:
For decades, there has been the de-facto standard method, EM(Expectation Maximization), for Gaussian Mixture Model(GMMs).
After briefly organizing EM-Algorithm, I will introduce as a potential replacement for EM Algorithm using a simple reformulation that makes Riemannian optimization not only match EM but also outperform it in many settings.