BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Gaussian Latent Tree Models and their Statistics. -Thomas Marge (s
 tatslab) - Thomas Marge (statslab)
DTSTART:20200307T112000Z
DTEND:20200307T115000Z
UID:TALK140740@talks.cam.ac.uk
CONTACT:73969
DESCRIPTION:Signal processing strategies and statistics for identifying th
 e presence of evolution in continuous signals is investigated. Consider a 
 feature to be a function on the original signal which contains information
  about the signal. Under this framework\, a model for multivariate Gaussia
 n features observed across related signals is described. The model conside
 rs the possibility that some features in the signal are tree amenable whil
 e others are not. A model for identifying candidate features using wavelet
  transforms is also described. Tree amenability is then explored from the 
 perspective of data thresholding. Because of the high type-1 and type-2 er
 ror rates of know tests for Gaussian tree amenability\, a measure of how t
 ree amenable a feature is has been developed. A methodology is proposed fo
 r reconstructing only the tree amenable components of a signal to improve 
 interoperability of the model. Rigorous statistical methods are then defin
 ed to test for both tree amenability as well as general structure in the d
 ata. To test and better understand these methods\, strategies are describe
 d to randomly generate tree amenable data.
LOCATION:Winstanley Lecture Theatre\, Trinity College
END:VEVENT
END:VCALENDAR