Date: May 6th, 2020. Kolkata time 8:00 pm.
SEMINAR SERIES ON RECENT PROGRESS IN GEOMETRIC COMPLEXITY THEORY
Meeting ID: 822 7078 2751
Gaussian group models
Kathlén Kohn and Carlos Améndola
KTH Stockholm and Technical University Munich.
The task of fitting data to a model is fundamental in statistics. For this, a widespread approach is maximum likelihood estimation (ML estimation), where one maximizes the likelihood of observing the data as we range over the model. In this talk we introduce Gaussian group models, which are multivariate Gaussian models parametrized by a representation of a group. In this general setting, we can characterize ML estimation by stability notions from invariant theory. Matrix normal models and Gaussian graphical models defined by transitive directed acyclic graphs (TDAGs) fit into this framework. For TDAGs, we provide exact conditions for the existence of the ML estimate in terms of linear independence of the rows of the sample matrix. This talk is based on joint work with Anna Seigal and Philipp Reichenbach.