BEGIN:VCALENDAR
PRODID:-//Columba Systems Ltd//NONSGML CPNG/SpringViewer/ICal Output/3.3-
M3//EN
VERSION:2.0
CALSCALE:GREGORIAN
METHOD:PUBLISH
BEGIN:VEVENT
DTSTAMP:20220925T181829Z
DTSTART:20221005T130000Z
DTEND:20221005T140000Z
SUMMARY:Suhasini Subba Rao - Graphical models for nonstationary time ser
ies (in person stat seminar)
UID:{http://www.columbasystems.com/customers/uom/gpp/eventid/}h7k-l7t586k
l-b6url4
DESCRIPTION:Suhasini Subba Rao\, Professor of Statistics in the Departmen
t of Statistics at Texas A&M University\, USA is our speaker for the Sta
tistics seminar series.\n\nTitle: Graphical models for nonstationary tim
e series\n\nAbstract: We propose NonStGM\, a general nonparametric grap
hical modeling framework for studying dynamic associations among the com
ponents of a nonstationary multivariate time series. It builds on the fr
amework of Gaussian Graphical Models (GGM) and stationary time series Gr
aphical models (StGM)\, and complements existing works on parametric gra
phical models based on change point vector autoregressions (VAR). Analog
ous to StGM\, the proposed framework captures conditional noncorrelation
s (both intertemporal and\ncontemporaneous) in the form of an undirected
graph. In addition\, to describe the more nuanced nonstationary relatio
nships among the components of the time series\, we introduce the new no
tion of conditional nonstationarity/stationarity and incorporate it with
in the graph architecture. \nThis allows one to distinguish between dir
ect and indirect nonstationary relationships among system components\, a
nd can be used to search for small subnetworks that serve as the ``sourc
e'' of nonstationarity in a large system. Together\, the two concepts of
conditional noncorrelation and nonstationarity/stationarity provide a
parsimonious description of the dependence structure of the time series.
\n\nIn GGM\, the graphical model structure is encoded in the sparsity pa
ttern of the inverse covariance matrix. Analogously\, we explicitly conn
ect conditional noncorrelation and stationarity between and within compo
nents of the multivariate time series to zero and\nToeplitz embeddings o
f an infinite-dimensional inverse covariance operator. In order to learn
the graph\, we move to the Fourier domain. We show that in the Fourier
domain\, conditional stationarity and noncorrelation relationships in th
e inverse covariance operator are encoded with a specific sparsity struc
ture of its integral kernel operator. Within the local stationary framew
ork we show that these sparsity patterns can be recovered from finite-le
ngth time series by node-wise regression of discrete Fourier Transforms
(DFT) across different Fourier frequencies.\nWe illustrate the features
of our general framework under the special case of time-varying Vector A
utoregressive models.
STATUS:TENTATIVE
TRANSP:TRANSPARENT
CLASS:PUBLIC
LOCATION:Frank Adams Seminar Room 2 \, Alan Turing Building \, Upper Broo
k street\, Manchester\, M13 9PL
END:VEVENT
END:VCALENDAR