BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Estimates for distributions of Hölder semi-norms of random proces
ses from spaces F_ψ(Ω)
DTSTART;VALUE=DATE-TIME:20170816T123000Z
DTEND;VALUE=DATE-TIME:20170816T130000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-270@indico.uu.se
DESCRIPTION:Speakers: Dmytro Zatula (Taras Shevchenko National University
of Kyiv)\n\nIn the following we deal with estimates for distributions of H
ölder semi-norms of sample functions of random processes from spaces $\\m
athbb{F}_\\psi(\\Omega)$\, defined on a compact metric space and on an inf
inite interval $[0\,\\infty)$\, i.e. probabilities\n$$\\mathsf{P}\\left\\{
\\sup\\limits_{\\substack{0x\\right\\}.$$\nSuch estimates and assumptions
under which semi-norms of sample functions of random processes from spaces
$\\mathbb{F}_\\psi(\\Omega)$\, defined on a compact space\, satisfy the H
ölder condition were obtained by Kozachenko and Zatula (2015). Similar re
sults were provided for Gaussian processes\, defined on a compact space\,
by Dudley (1973). Kozachenko (1985) generalized Dudley's results for rando
m processes belonging to Orlicz spaces\, see also Buldygin and Kozachenko
(2000). Marcus and Rosen (2008) obtained $L^p$ moduli of continuity for a
wide class of continuous Gaussian processes. Kozachenko et al. (2011) stud
ied the Lipschitz continuity of generalized sub-Gaussian processes and pro
vided estimates for the distribution of Lipschitz norms of such processes.
But all these problems were not considered yet for processes\, defined on
an infinite interval.\n\nhttps://indico.uu.se/event/317/contributions/270
/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/270/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Modeling of vertical and horizontal variation in multivariate func
tional data
DTSTART;VALUE=DATE-TIME:20170815T140000Z
DTEND;VALUE=DATE-TIME:20170815T143000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-271@indico.uu.se
DESCRIPTION:Speakers: Niels Olsen (Københvans Universitet)\n\nWe present
a model for multivariate functional data that simultaneously model vertica
l and horisontal variation. \nHorisontal variation is modeled using warpin
g functions represented by latent gaussian variables.\nVertical variation
is modeled using Gaussian processes using a generally applicable low-param
etric covariance structure.\nWe devise a method for maximum likelihood es
timation using a Laplace approximation and apply it to three different dat
a sets.\n\nhttps://indico.uu.se/event/317/contributions/271/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/271/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Finite Mixture of C-vines for Complex Dependence
DTSTART;VALUE=DATE-TIME:20170817T070000Z
DTEND;VALUE=DATE-TIME:20170817T073000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-272@indico.uu.se
DESCRIPTION:Speakers: A. Sevtap Kestel (Middle East Technical University)\
, Ceylan Yozgatlıgil (Middle East Technical University)\, O. Ozan Evkaya
(Atılım University)\n\nRecently\, there has been an increasing interest
on the combination of copulas with a finite mixture model. Such a framewor
k is useful to reveal the hidden dependence patterns observed for random v
ariables flexibly in terms of statistical modeling. The combination of vin
e copulas incorporated into a finite mixture model is also beneficial for
capturing hidden structures on a multivariate data set. In this respect\,
the main goal of this study is extending the study of Kim et al. (2013) wi
th different scenarios. For this reason\, finite mixture of C-vine is prop
osed for multivariate data with different dependence structures. The perfo
rmance of the proposed model has been tested by different simulated data s
et including various tail dependence properties.\n\nhttps://indico.uu.se/e
vent/317/contributions/272/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/272/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Testing independence for multivariate time series by the auto-dist
ance correlation matrix
DTSTART;VALUE=DATE-TIME:20170815T093000Z
DTEND;VALUE=DATE-TIME:20170815T100000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-273@indico.uu.se
DESCRIPTION:Speakers: Maria Pitsillou (Department of Mathematics &\; St
atistics\, Cyprus)\, Konstantinos Fokianos (Department of Mathematics & St
atistics\, University of Cyprus)\n\nWe introduce the notions of multivaria
te auto-distance covariance and correlation functions\nfor time series ana
lysis. These concepts have been recently discussed in the context\nof both
independent and dependent data but we extend them in a different directio
n by\nputting forward their matrix version. Their matrix version allows us
to identify possible\ninterrelationships among the components of a multiv
ariate time series. Interpretation\nand consistent estimators of these new
concepts are discussed. Additionally\, we develop\na test for testing the
i.i.d. hypothesis for multivariate time series data. The resulting test\n
statistic performs better than the standard multivariate Ljung-Box test st
atistic. All the\nabove methodology is included in the R package dCovTS wh
ich is briefly introduced in\nthis talk.\n\nhttps://indico.uu.se/event/317
/contributions/273/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/273/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Theoretical and simulation results on heavy-tailed fractional Pear
son diffusions
DTSTART;VALUE=DATE-TIME:20170818T083000Z
DTEND;VALUE=DATE-TIME:20170818T090000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-274@indico.uu.se
DESCRIPTION:Speakers: Nenad Šuvak (Department of Mathematics\, J.J. Stros
smayer University of Osijek)\, Alla Sikorskii (Department of Statistics an
d Probability\, Michigan State University)\, Nikolai N. Leonenko (School o
f Mathematics\, Cardiff University)\, Ivan Papić (Department of Mathemati
cs\, J.J. Strossmayer University of Osijek)\n\nWe define heavy-tailed frac
tional reciprocal gamma and Fisher-Snedecor diffusions by a non-Markovian
time change in the corresponding Pearson diffusions. We illustrate known t
heoretical results regarding these fractional diffusions via simulations.\
n\nhttps://indico.uu.se/event/317/contributions/274/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/274/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Copula based BINAR models with applications
DTSTART;VALUE=DATE-TIME:20170818T090000Z
DTEND;VALUE=DATE-TIME:20170818T093000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-275@indico.uu.se
DESCRIPTION:Speakers: Andrius Buteikis (Faculty of Mathematics and Informa
tics\, Vilnius University)\n\nIn this paper we study the problem of modell
ing the integer-valued vector observations. We consider the BINAR(1) model
s defined via copula-joint innovations. We review different parameter esti
mation methods and analyse estimation methods of the copula dependence par
ameter. We also examine the case where seasonality is present in integer-v
alued data and suggest a method of deseasonalizing them. Finally\, an empi
rical application is carried out.\n\nhttps://indico.uu.se/event/317/contri
butions/275/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/275/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Fréchet means and Procrustes analysis in Wasserstein space
DTSTART;VALUE=DATE-TIME:20170815T133000Z
DTEND;VALUE=DATE-TIME:20170815T140000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-276@indico.uu.se
DESCRIPTION:Speakers: Yoav Zemel (Ecole polytechnique fédérale de Lausan
ne)\, Victor Panaretos (Ecole polytechnique fédérale de Lausanne)\n\nWe
consider three interlinked problems in stochastic geometry: (1) construct
ing optimal multicouplings of random vectors\; (2) determining the Fréch
et mean of probability measures in Wasserstein space\; and (3) registering
collections of randomly deformed spatial point processes. We demonstrate
how these problems are canonically interpreted through the prism of the t
heory of optimal transportation of measure on $\\mathbb R^d$. We provide
explicit solutions in the one dimensional case\, consistently solve the re
gistration problem and establish convergence rates and a (tangent space) c
entral limit theorem for Cox processes. When $d>1$\, the solutions are no
longer explicit and we propose a steepest descent algorithm for deducing
the Fréchet mean in problem (2). Supplemented by uniform convergence res
ults for the optimal maps\, this furnishes a solution to the multicoupling
problem (1). The latter is then utilised\, as in the case $d=1$\, in ord
er to construct consistent estimators for the registration problem (3). W
hile the consistency results parallel their one-dimensional counterparts\,
their derivation requires more sophisticated techniques from convex analy
sis. This is joint work with Victor M. Panaretos\n\nhttps://indico.uu.se/
event/317/contributions/276/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/276/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Efficient estimation for diffusions
DTSTART;VALUE=DATE-TIME:20170816T120000Z
DTEND;VALUE=DATE-TIME:20170816T123000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-277@indico.uu.se
DESCRIPTION:Speakers: Nina Munkholt Jakobsen (University of Copenhagen)\,
Michael Sørensen (University of Copenhagen)\n\nThis talk concerns estimat
ion of the diffusion parameter of a diffusion process observed over a fixe
d time interval. We present conditions on approximate martingale estimatin
g functions under which estimators are consistent\, rate optimal\, and eff
icient under high frequency (in-fill) asymptotics. Here\, limit distributi
ons of the estimators are non-standard in the sense that they are generall
y normal variance-mixture distributions. In particular\, the mixing distri
bution depends on the full sample path of the diffusion process over the o
bservation time interval. Making use of stable convergence in distribution
\, we also present the more easily applicable result that estimators norma
lized by a suitable data-dependent transformation converge in distribution
to a standard normal distribution. The theory is illustrated by a simulat
ion study.\n\nThe work presented in this talk is published in:\n\nJakobsen
\, N. M. and Sørensen\, M. (2017). *Efficient estimation for diffusions s
ampled at high frequency over a fixed time interval.* Bernoulli\, 23(3):18
74-1910.\n\nhttps://indico.uu.se/event/317/contributions/277/
LOCATION:Museum Gustavianum (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/277/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Best Unbiased Estimators for Doubly Multivariate Data
DTSTART;VALUE=DATE-TIME:20170815T143000Z
DTEND;VALUE=DATE-TIME:20170815T150000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-278@indico.uu.se
DESCRIPTION:Speakers: Anuradha Roy (Department of Management Science and S
tatistics The University of Texas at San Antonio San Antonio\, TX 78249\,
USA)\, Miguel Fonseca (Centro de Matemática e Aplicações Universidade N
ova de Lisboa Monte da Caparica\, 2829-516 Caparica\, Portugal)\, Ricardo
Leiva (Departamento de Matemática F.C.E.\, Universidad Nacional de Cuyo\,
5500 Mendoza\, Argentina)\, Roman Zmyślony (Faculty of Mathematics\, Com
puter Science and Econometrics University of Zielona Góra\, Szafrana 4a\,
65-516 Zielona Góra\, Poland)\, Arkadiusz Kozioł (Faculty of Mathematic
s\, Computer Science and Econometrics University of Zielona Góra\, Szafra
na 4a\, 65-516 Zielona Góra\, Poland)\n\nThe article addresses the best u
nbiased estimators of the block compound symmetric covariance\nstructure f
or m-variate observations with equal mean vector over each level of factor
or each time point (model with structured mean vector). Under multivariat
e normality\, the free-coordinate approach is used to obtain unbiased line
ar and quadratic estimates for the model parameters. Optimality of these e
stimators follows from sufficiency and completeness of their distributions
. Additionally\, strong consistency is proven. The properties of the estim
ators in the proposed model are compared with the ones in the model with u
nstructured mean vector (the mean vector changes over levels of factor or
time points).\n\nhttps://indico.uu.se/event/317/contributions/278/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/278/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Stability of the Spectral EnKF under nested covariance estimators
DTSTART;VALUE=DATE-TIME:20170817T143000Z
DTEND;VALUE=DATE-TIME:20170817T150000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-279@indico.uu.se
DESCRIPTION:Speakers: Krystof Eben (Institute of Computer Science\, The Cz
ech Academy of Sciences)\, Jan Mandel (University of Colorado Denver)\, Ma
rie Turčičová (Charles University\, Prague)\n\nIn the case of tradition
al Ensemble Kalman Filter (EnKF)\, it is known that the filter error does
not \ngrow faster than exponentially for a fixed ensemble size. The questi
on posted in this contribution is whether the upper bound for the filter e
rror can be improved by using an improved covariance estimator that comes
from the right parameter subspace and has smaller asymptotic variance. Its
effect on Spectral EnKF is explored by a simulation.\n\nhttps://indico.uu
.se/event/317/contributions/279/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/279/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Invited Speaker - Random Networks
DTSTART;VALUE=DATE-TIME:20170818T070000Z
DTEND;VALUE=DATE-TIME:20170818T080000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-280@indico.uu.se
DESCRIPTION:Speakers: Svante Janson (Uppsala University)\n\nhttps://indico
.uu.se/event/317/contributions/280/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/280/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Matrix Independent Component Analysis
DTSTART;VALUE=DATE-TIME:20170814T133000Z
DTEND;VALUE=DATE-TIME:20170814T140000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-281@indico.uu.se
DESCRIPTION:Speakers: Joni Virta (University of Turku)\n\nIndependent comp
onent analysis (ICA) is a popular means of dimension reduction for vector-
valued random variables. In this short note we review its extension to arb
itrary tensor-valued random variables by considering the special case of t
wo dimensions where the tensors are simply matrices.\n\nhttps://indico.uu.
se/event/317/contributions/281/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/281/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Can humans be replaced by computers in taxa recognition?
DTSTART;VALUE=DATE-TIME:20170814T120000Z
DTEND;VALUE=DATE-TIME:20170814T123000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-282@indico.uu.se
DESCRIPTION:Speakers: Salme Kärkkäinen (Department of Mathematics and St
atistics\, University of Jyväskylä)\, Kristian Meissner (Finnish Environ
ment Institute)\, Jenni Raitoharju (Department of Signal Processing\, Tamp
ere University of Technology)\, Ville Tirronen (Department of Mathematical
Information Technology\, University of Jyväskylä)\, Johanna Ärje (Univ
ersity of Jyväskylä\, Department of Mathematics and Statistics)\n\nBiomo
nitoring of waterbodies is vital as the number of anthropogenic stressors
on aquatic ecosystems keeps growing. However\, the continuous decrease in
funding makes it impossible to meet monitoring goals or sustain traditiona
l manual sample processing. We review what kind of statistical tools can b
e used to enhance the cost efficiency of biomonitoring: We explore automat
ed identification of freshwater macroinvertebrates which are used as one i
ndicator group in biomonitoring of aquatic ecosystems. We present the firs
t classification results of a new imaging system producing multiple images
per specimen. Moreover\, these results are compared with the results of h
uman experts. On a data set of 29 taxonomical groups\, automated classific
ation produces a higher average accuracy than human experts.\n\nhttps://in
dico.uu.se/event/317/contributions/282/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/282/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Nonparametric estimation of gradual change points in the jump beha
viour of an Ito semimartingale
DTSTART;VALUE=DATE-TIME:20170814T123000Z
DTEND;VALUE=DATE-TIME:20170814T130000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-283@indico.uu.se
DESCRIPTION:Speakers: Holger Dette (Ruhr-Universität Bochum)\, Mathias Ve
tter (Christian-Albrechts-Universität zu Kiel)\, Michael Hoffmann (Ruhr-U
niversität Bochum)\n\nIn applications the properties of a stochastic feat
ure often change gradually rather than\nabruptly\, that is: after a consta
nt phase for some time they slowly start to vary. The goal of this talk is
to introduce an estimator for the location of a gradual change point in t
he jump characteristic of a\ndiscretely observed Ito semimartingale. To th
is end we propose a measure of time variation for the\njump behaviour of t
he process and consistency of the desired estimator is a consequence of we
ak convergence of a suitable empirical process in some function space. Fin
ally\, we discuss simulation results which verify that the new estimator h
as advantages compared to the classical argmax-estimator.\n\nhttps://indic
o.uu.se/event/317/contributions/283/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/283/
END:VEVENT
BEGIN:VEVENT
SUMMARY:AIC post-selection inference in linear regression
DTSTART;VALUE=DATE-TIME:20170814T140000Z
DTEND;VALUE=DATE-TIME:20170814T143000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-284@indico.uu.se
DESCRIPTION:Speakers: Ali Charkhi (KULeuven)\, Gerda Claeskens (KULeuven)\
n\nPost-selection inference has been considered a crucial topic in data\na
nalysis. In this article\, we develop a new method to obtain correct infer
ence after model selection by the Akaike's information criterion Akaike (1
973) in linear regression models. Confidence intervals can be calculated b
y incorporating the randomness of the model selection in the distribution
of the parameter estimators which act as pivotal quantities. Simulation re
sults show the accuracy of the proposed method.\n\nhttps://indico.uu.se/ev
ent/317/contributions/284/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/284/
END:VEVENT
BEGIN:VEVENT
SUMMARY:E-optimal approximate block designs for treatment-control comparis
ons
DTSTART;VALUE=DATE-TIME:20170817T120000Z
DTEND;VALUE=DATE-TIME:20170817T123000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-298@indico.uu.se
DESCRIPTION:Speakers: Samuel Rosa (Comenius University in Bratislava)\n\nW
e study $E$-optimal block designs for comparing a set of test treatments w
ith a control treatment. We provide the complete class of all $E$-optimal
approximate block designs and we show that these designs are characterized
by simple linear constraints. Employing the provided characterization\, w
e obtain a class of $E$-optimal exact block designs with unequal block siz
es for comparing test treatments with a control.\n\nhttps://indico.uu.se/e
vent/317/contributions/298/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/298/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Viterbi process for pairwise Markov models
DTSTART;VALUE=DATE-TIME:20170817T080000Z
DTEND;VALUE=DATE-TIME:20170817T083000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-299@indico.uu.se
DESCRIPTION:Speakers: Joonas Sova (University of Tartu)\n\nMy talk is base
d on ongoing joint work with my supervisor Jüri Lember.\n\nWe consider a
Markov chain $Z = \\{Z_k\\}_{k \\geq 1}$ with product\nstate space $\\math
cal{X}\\times \\mathcal{Y}$\, where $\\mathcal{Y}$ is\na finite set (state
space) and $\\mathcal{X}$ is an arbitrary\nseparable metric space (observ
ation space). Thus\, the process $Z$\ndecomposes as $Z=(X\,Y)$\, where $X=
\\{X_k \\}_{k\\geq 1}$ and $Y=\\{Y_k\n\\}_{k\\geq 1}$ are random processes
taking values in $\\mathcal{X}$\nand $\\mathcal{Y}$\, respectively. Foll
owing\ncite{pairwise\,pairwise2\,pairwise3}\, we call the process $Z$ a\n\
\textit{pairwise Markov model}. The process $X$ is identified as an\nobser
vation process and the process $Y$\, sometimes called the \\textit{regime}
\, models the observations-driving hidden state sequence.\nTherefore our
general model contains many well-known stochastic\nmodels as a special cas
e: hidden Markov models\, Markov\nswitching models\, hidden Markov models
with dependent noise and many\nmore. The \\textit{segmentation} or \\text
it{path estimation} problem\nconsists of estimating the realization of $(Y
_1\,\\ldots\,Y_n)$ given a\nrealization $x_{1:n}$ of $(X_1\,\\ldots\,X_n)$
. A standard estimate is\nany path $v_{1:n}\\in \\mathcal{Y}^n$ having max
imum posterior\nprobability:\n$$v_{1:n}=\\mathop{\\mathrm{argmax}}_{y_{1:n
}}P(Y_{1:n}=y_{1:n}|X_{1:n}=x_{1:n}).$$\nAny such path is called \\textit
{Viterbi path} and we are interested in\nthe behaviour of $v_{1:n}$ as $n$
grows. The study of asymptotics of\nViterbi path is complicated by the fa
ct that adding one more\nobservation\, $x_{n+1}$ can change the whole path
\, and so it is not\nclear\, whether there exists a limiting infinite Vite
rbi path. \n\nWe show that under some conditions the infinite Viterbi path
indeed exists \nfor almost every realization $x_{1:\\infty}$ of $X$\, the
reby defining an infinite Viterbi decoding of $X$\, called the \\textit{Vi
terbi process.} This is done trough construction of \\textit{barriers}. A
barrier is a fixed-sized block in the observations $x_{1:n}$ that fixes t
he Viterbi path up to\n itself: for every continuation of $x_{1:n}$\, the
Viterbi path up to\n the barrier remains unchanged. Therefore\, if\nalmost
every realization of $X$-process contains\ninfinitely many barriers\, the
n the Viterbi process exists.\n\nHaving infinitely many barriers is not ne
cessary for\nexistence of infinite Viterbi path\, but the\nbarrier-constru
ction has several advantages. One of them is that it\nallows to construct
the infinite path \\textit{piecewise}\, meaning that\nto determine the fir
st $k$ elements $v_{1:k}$ of the infinite path\nit suffices to observe $x_
{1:n}$ for $n$ big enough. Barrier construction has another great advantag
e: namely\, the process $(Z\,V)=\\{(Z_k\,V_k)\\}_{k \\geq 1}$\, where $V=
\\{V_k\\}_{k \\geq 1}$ denotes the Viterbi process\, is under certain cond
itions regenerative. This is can be proven by\, roughly speaking\, applyin
g the Markov splitting method to construct regeneration times for $Z$ whic
h coincide with the occurrences of barriers. Regenerativity of $(Z\,V)$ al
lows to easily prove limit theorems to understand the asymptotic behaviour
of inferences based on Viterbi\npaths. In fact\, in a special case of hid
den Markov model this regenerative property has already been known to hold
and has found several applications cite{AV\,AVacta\,Vsmoothing\,Vrisk\, i
owa}.\n\nhttps://indico.uu.se/event/317/contributions/299/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/299/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Predict extreme influenza epidemics
DTSTART;VALUE=DATE-TIME:20170816T073000Z
DTEND;VALUE=DATE-TIME:20170816T080000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-286@indico.uu.se
DESCRIPTION:Speakers: Maud Thomas (Université Pierre et Marie Curie)\, Ho
lger Rootzén (Chalmers University of Technology)\n\nInfluenza viruses are
responsible for annual epidemics\, causing more than 500\,000 deaths per
year worldwide. A crucial question for resource planning in public health
is to predict the morbidity burden of extreme epidemics. We say that an ep
idemic is extreme whenever the influenza incidence rate exceeds a high thr
eshold for at least one week. Our objective is to predict whether an extre
me epidemic will occur in the near future\, say the next couple of weeks.\
n\nThe weekly numbers of influenza-like illness (ILI) incidence rates in F
rance are available from the Sentinel network for the period 1991-2017. IL
I incidence rates exhibit two different regimes\, an epidemic regime durin
g winter and a non-epidemic regime during the rest of the year. To identif
y epidemic periods\, we use a two-state autoregressive hidden Markov model
.\n\nA main goal of Extreme Value Theory is to assess\, from a series of o
bservations\, the probability of events that are more extreme than those p
reviously recorded. Because of the autoregressive structure of the data\,
we choose to fit one of the mul- tivariate generalized Pareto distribution
models proposed in Rootzén et al. (2016a) [Multivariate peaks over thres
hold models. arXiv:1603.06619v2]\; see also Rootzén et al. (2016b) [Peaks
over thresholds modeling with multivariate generalized Pareto distributio
ns. arXiv:1612.01773v1]. For these models\, explicit densities are given\,
and formulas for conditional probabilities can then be deduced\, from whi
ch we can predict if an epidemic will be extreme\, given the first weeks o
f observation.\n\nhttps://indico.uu.se/event/317/contributions/286/
LOCATION:Museum Gustavianum (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/286/
END:VEVENT
BEGIN:VEVENT
SUMMARY:The Elicitation Problem
DTSTART;VALUE=DATE-TIME:20170815T090000Z
DTEND;VALUE=DATE-TIME:20170815T093000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-287@indico.uu.se
DESCRIPTION:Speakers: Tobias Fissler (University of Bern)\n\nCompeting poi
nt forecasts for functionals such as the mean\, a quantile\, or a certain
risk measure are commonly compared in terms of loss functions. These shoul
d be incentive compatible\, i.e.\, the expected score should be minimized
by the correctly specified functional of interest. A functional is called
*elicitable* if it possesses such an incentive compatible loss function. W
ith the squared loss and the absolute loss\, the mean and the median posse
ss such incentive compatible loss functions\, which means they are elicita
ble. In contrast\, variance or Expected Shortfall are not elicitable. \nBe
sides investigating the elicitability of a functional\, it is important to
determine the whole class of incentive compatible loss functions as well
as to give recommendations which loss function to use in practice\, taking
into regard secondary quality criteria of loss functions such as order-se
nsitivity\, convexity\, or homogeneity.\n\nhttps://indico.uu.se/event/317/
contributions/287/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/287/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Controlled branching processes in Biology: a model for cell prolif
eration
DTSTART;VALUE=DATE-TIME:20170815T120000Z
DTEND;VALUE=DATE-TIME:20170815T123000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-288@indico.uu.se
DESCRIPTION:Speakers: Inés María del Puerto García (University of Extre
madura)\, Miguel González Velasco (University of Extremadura)\, Carmen Mi
nuesa Abril (University of Extremadura)\n\nBranching processes are relevan
t models in the development of theoretical approaches to problems in appli
ed fields such as\, for instance\, growth and extinction of populations\,
biology\, epidemiology\, cell proliferation kinetics\, genetics and algori
thm and data structures. The most basic model\, the so-called Bienaymé-Ga
lton-Watson process\, consists of individuals that reproduce independently
of the others following the same probability distribution\, known as offs
pring distribution. A natural generalization is to incorporate a random co
ntrol function which determines the number of progenitors in each generati
on. The resulting process is called controlled branching process.\n\nIn th
is talk\, we deal with a problem arising in cell biology. More specificall
y\, we focus our attention on experimental data generated by time-lapse vi
deo recording of cultured in vitro oligodendrocyte cells. In A.Y. Yakovlev
et al. (2008) (Branching Processes as Models of Progenitor Cell Populatio
ns and Estimation of the Offspring Distributions\, *Journal of the America
n Statistical Association*\, 103(484):1357--1366)\, a two-type age depende
nt branching process with emigration is considered to describe the kinetic
s of cell populations. The two types of cells considered are referred as
type $T_1$ (immediate precursors of oligodendrocytes) and type $T_2$ (term
inally differentiated oligodendrocytes). The reproduction process of these
cells is as follows: when stimulating to divide under in vitro conditions
\, the progenitor cells are capable of producing either their direct proge
ny (two daughter cells of the same type) or a single\, terminally differen
tiated nondividing oligodendrocyte. Moreover\, censoring effects as a cons
equence of the migration of progenitor cells out of the microscopic field
of observation are modelled as the process of emigration of the type $T_1$
cells.\n\nIn this work\, we propose a two-type controlled branching proce
ss to describe the embedded discrete branching structure of the age-depend
ent branching process aforementioned. We address the estimation of the off
spring distribution of the cell population in a Bayesian outlook by making
use of disparities. The importance of this problem yields in the fact tha
t the behaviour of these populations is strongly related to the main param
eters of the offspring distribution and in practice\, these values are unk
nown and their estimation is necessary. The proposed methodology introduce
d in M. Gonz\\'alez et al. (2017) (Robust estimation in controlled branchi
ng processes: Bayesian estimators via disparities. *Work in progress*)\, i
s illustrated with an application to the real data set given in A.Y. Yakov
lev et al. (2008).\n\nhttps://indico.uu.se/event/317/contributions/288/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/288/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Invited Speaker - Independent component analysis using third and f
ourth cumulants
DTSTART;VALUE=DATE-TIME:20170817T110000Z
DTEND;VALUE=DATE-TIME:20170817T120000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-290@indico.uu.se
DESCRIPTION:Speakers: Hannu Oja (University of Turku)\n\nIn independent co
mponent analysis it is assumed that the observed random variables are line
ar combinations of latent\, mutually independent random variables called t
he independent components. It is then often thought that only the non-Gaus
sian independent components are of interest and the Gaussian components si
mply present noise. The idea is then to make inference on the unknown numb
er of non-Gaussian components and to estimate the transformations back to
the non-Gaussian components.\n \nIn this talk we show how the classical sk
ewness and kurtosis measures\, namely third and fourth cumulants\, can be
used in the estimation. First\, univariate cumulants are used as projectio
n indices \nin search for independent components (projection pursuit\, fas
tICA). Second\, multivariate fourth cumulant matrices are jointly used to
solve the problem (FOBI\, JADE). The properties of the estimates are consi
dered through corresponding optimization problems\, estimating equations\,
algorithms and asymptotic statistical properties. The theory is illustrat
ed with several examples.\n\nhttps://indico.uu.se/event/317/contributions/
290/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/290/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Simulating and Forecasting Human Population with General Branching
Process
DTSTART;VALUE=DATE-TIME:20170818T093000Z
DTEND;VALUE=DATE-TIME:20170818T100000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-300@indico.uu.se
DESCRIPTION:Speakers: Plamen Trayanov (Sofia Univeristy "St. Kliment Ohrid
ski")\n\nThe branching process theory is widely used to describe a populat
ion dynamics in which particles live and produce other particles through t
heir life\, according to given stochastic birth and death laws. The theory
of General Branching Processes (GBP) presents a continuous time model in
which every woman has random life length and gives birth to children in ra
ndom intervals of time. The flexibility of the GBP makes it very useful fo
r modelling and forecasting human population. This paper is a continuation
of previous developments in the theory\, necessary to model the specifics
of human population\, and presents their application in forecasting the p
opulation age structure of Bulgaria. It also introduces confidence interva
ls of the forecasts\, calculated by GBP simulations\, which reflect both t
he stochastic nature of the birth and death laws and the branching process
itself. The simulations are also used to determine the main sources of ri
sk to the forecast.\n\nhttps://indico.uu.se/event/317/contributions/300/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/300/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Parameter Estimation for Discretely Observed Infinite-Server Queue
s with Markov-Modulated Input
DTSTART;VALUE=DATE-TIME:20170815T123000Z
DTEND;VALUE=DATE-TIME:20170815T130000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-291@indico.uu.se
DESCRIPTION:Speakers: Michel Mandjes (Universiteit van Amsterdam)\, Bartek
Knapik (Vrije Universiteit Amsterdam)\, Mathisca de Gunst (Vrije Universi
teit Amsterdam)\, Birgit Sollie (Vrije Universiteit Amsterdam)\n\nThe Mark
ov-modulated infinite-server queue is a queueing system with infinitely ma
ny servers\, where the arrivals follow a Markov-modulated Poisson process
(MMPP)\, i.e. a Poisson process with rate modulating between several value
s. The modulation is driven by an underlying and unobserved continuous tim
e Markov chain $\\{X_t\\}_{t\\geq 0}$. The inhomogeneous rate of the Poiss
on process\, $\\lambda(t)$\, stochastically alternates between $d$ differe
nt rates\, $\\lambda_1\,\\dots\,\\lambda_d$\, in such a way that $\\lambda
(t) = \\lambda_i$ if $X_t = i$\, $i=1\,\\dots\,d$.\n\nWe are interested in
estimating the parameters of the arrival process for this queueing system
based on observations of the queue length at discrete times only. We assu
me exponentially distributed service times with rate $\\mu$\, where $\\mu$
is time-independent and known. Estimation of the parameters of the arriva
l process has not yet been studied for this particular queueing system. Tw
o types of missing data are intrinsic to the model\, which complicates the
estimation problem. First\, the underlying continuous time Markov chain i
n the Markov-modulated arrival process is not observed. Second\, the queue
length is only observed at a finite number of discrete time points. As a
result\, it is not possible to distinguish the number of arrivals and the
number of departures between two consecutive observations. \n\nIn this tal
k we show how we derive an explicit algorithm to find maximum likelihood e
stimates of the parameters of the arrival process\, making use of the EM a
lgorithm. Our approach extends the one used in Okamura et al. (2009)\, whe
re the parameters of an MMPP are estimated based on observations of the p
rocess at discrete times. However\, in contrast to our setting\, Okamura e
t al. (2009) do not consider departures and therefore do not deal with the
second type of missing data. We illustrate the accuracy of the proposed e
stimation algorithm with a simulation study.\n\nReference: Okamura H.\, Do
hi T.\, Trivedi K.S.\n(2009).\nMarkovian Arrival Process Parameter Estimat
ion With Group Data.\nIEEE/ACM Transactions on Networking. \nVol. 17\, No.
4\, pp. 1326--1339\n\nhttps://indico.uu.se/event/317/contributions/291/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/291/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Mallows' Model Based on Lee Distance
DTSTART;VALUE=DATE-TIME:20170817T133000Z
DTEND;VALUE=DATE-TIME:20170817T140000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-292@indico.uu.se
DESCRIPTION:Speakers: Nikolay Nikolov (Institute of Mathematics and Inform
atics\, Bulgarian Academy of Sciences\, Acad. G.Bontchev str.\, block 8\,
1113 Sofia\, Bulgaria)\, Eugenia Stoimenova (Institute of Information and
Communication Technologies and Institute of Mathematics and Informatics\,
Bulgarian Academy of Sciences\, Acad. G.Bontchev str.\, block 25A\, 1113 S
ofia\, Bulgaria)\n\nIn this paper the Mallows' model based on Lee distance
is considered and compared to models induced by other metrics on the perm
utation group. As an illustration\, the complete rankings from the America
n Psychological Association election data are analyzed.\n\nhttps://indico.
uu.se/event/317/contributions/292/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/292/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Multilevel Functional Principal Component Analysis for Unbalanced
Data
DTSTART;VALUE=DATE-TIME:20170814T143000Z
DTEND;VALUE=DATE-TIME:20170814T150000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-293@indico.uu.se
DESCRIPTION:Speakers: Zuzana Rošťáková (Institute of Measurement Scien
ce\, Slovak Academy of Sciences)\n\nFunctional principal component analysi
s (FPCA) is the key technique for dimensionality reduction and detection o
f main directions of variability present in functional data. However\, it
is not the most suitable tool for the situation when analyzed dataset cont
ains repeated or multiple observations\, because information about repeat
ability of measurements is not taken into account. Multilevel functional p
rincipal component analysis (MFPCA) is the modified version of FPCA develo
ped for data observed at multiple visits. The original MFPCA method was de
signed for balanced data only\, where for each subject the same number of
measurements is available. In this article we propose the modified MFPCA a
lgorithm which can be applied for unbalanced functional data\; that is\, i
n the situation where a different number of observations can be present fo
r every subject. The modified algorithm is validated and tested on real-wo
rld sleep data.\n\nhttps://indico.uu.se/event/317/contributions/293/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/293/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Some recent characterization based goodness of fit tests
DTSTART;VALUE=DATE-TIME:20170815T110000Z
DTEND;VALUE=DATE-TIME:20170815T113000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-294@indico.uu.se
DESCRIPTION:Speakers: Bojana Milošević (Faculty of Mathematics)\n\nIn th
is paper some recent advances in goodness of fit testing are presented.
Special attention is given to goodness of fit tests based on equidistribut
ion and independence characterizations. New concepts are described through
some modern exponentiality tests. Their natural generalizations are also
proposed. All tests are compared in Bahadur sense.\n\nhttps://indico.uu.se
/event/317/contributions/294/
LOCATION:Ångströmslaboratoriet
URL:https://indico.uu.se/event/317/contributions/294/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Invited Speaker - Independent component analysis using third and f
ourth cumulants
DTSTART;VALUE=DATE-TIME:20170817T090000Z
DTEND;VALUE=DATE-TIME:20170817T100000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-295@indico.uu.se
DESCRIPTION:Speakers: Hannu Oja (University of Turku)\n\nIn independent co
mponent analysis it is assumed that the observed random variables are line
ar combinations of latent\, mutually independent random variables called t
he independent components. It is then often thought that only the non-Gaus
sian independent components are of interest and the Gaussian components si
mply present noise. The idea is then to make inference on the unknown numb
er of non-Gaussian components and to estimate the transformations back to
the non-Gaussian components.\n \nIn this talk we show how the classical sk
ewness and kurtosis measures\, namely third and fourth cumulants\, can be
used in the estimation. First\, univariate cumulants are used as projectio
n indices \nin search for independent components (projection pursuit\, fas
tICA). Second\, multivariate fourth cumulant matrices are jointly used to
solve the problem (FOBI\, JADE). The properties of the estimates are consi
dered through corresponding optimization problems\, estimating equations\,
algorithms and asymptotic statistical properties. The theory is illustrat
ed with several examples.\n\nhttps://indico.uu.se/event/317/contributions/
295/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/295/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Joint Bayesian nonparametric reconstruction of dynamical equations
DTSTART;VALUE=DATE-TIME:20170817T073000Z
DTEND;VALUE=DATE-TIME:20170817T080000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-296@indico.uu.se
DESCRIPTION:Speakers: Christos Merkatas (Department of Mathematics\, Unive
rsity of the Aegean\, Greece)\, Spyridon Hatjispyros (Department of Mathem
atics\, University of the Aegean\, Greece)\n\nWe propose a Bayesian nonpar
ametric mixture model for the joint full reconstruction of $m$ dynamical e
quations\, \ngiven $m$ observed dynamically-noisy-corrupted chaotic time s
eries. The method of reconstruction is based on the Pairwise Dependent Geo
metric Stick Breaking Processes mixture priors (PDGSBP) first proposed by
Hatjispyros et al. (2017). We assume that\neach set of dynamical equations
has a deterministic part with a known functional form i.e. \n$$\nx_{ji} =
g_{j}(\\vartheta_j\, x_{j\,i-1}\,\\ldots\,x_{j\,i-l_j}) + \\epsilon_{x_{j
i}}\,\\\,\\\,\\\, 1\\leq j \\leq m\,\\\,\\\, 1\\leq i \\leq n_{j}.\n$$\nun
der the assumption that the noise processes $(\\epsilon_{x_{ji}})$ are ind
ependent and identically distributed for all $j$ and $i$ from some unknown
zero mean process $f_j(\\cdot)$. Additionally\, we assume that a-priori w
e have the knowledge that the processes $(\\epsilon_{x_{ji}})$ for $j=1\,\
\ldots\,m$ have common characteristics\,\ne.g. they may have common varian
ces or even have similar tail behavior etc. For a full reconstruction\, we
would like to jointly estimate the following quantities\n$$\n(\\vartheta_
{j})\\in\\Theta\\subseteq{\\cal R}^{k_j}\,\\quad (x_{j\,0}\,\\ldots\, x_{j
\,l_j-1})\\in{\\cal X}_j\\subseteq{\\cal R}^{l_j}\,\n$$\nand perform densi
ty estimation to the $m$ noise components $(f_j)$.\n\nOur contention is th
at whenever there is at least one sufficiently large data set\, using care
fully selected informative \nborrowing-of-strength-prior-specifications w
e are able to reconstruct those dynamical processes that are responsible f
or \nthe generation of time series with small sample sizes\; namely sample
sizes that are inadequate for an independent reconstruction.\nWe illustra
te the joint estimation process for the case $m=2$\, when the two time ser
ies are coming from a quadratic and a cubic \nstochastic process of lag on
e and the noise processes are zero mean normal mixtures with common compon
ents.\n\nhttps://indico.uu.se/event/317/contributions/296/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/296/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Confidence regions in Cox proportional hazards model with measurem
ent errors
DTSTART;VALUE=DATE-TIME:20170817T140000Z
DTEND;VALUE=DATE-TIME:20170817T143000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-297@indico.uu.se
DESCRIPTION:Speakers: Oksana Chernova (Taras Shevchenko National Universit
y of Kyiv)\n\nCox proportional hazards model with measurement errors in co
variates is considered. It is the ubiquitous technique in biomedical data
analysis. In Kukush et al. (2011) [ Journal of Statistical Research **45
**\, 77-94 ] and Chimisov and Kukush (2014) [ Modern Stochastics: Theory
and Applications **1**\, 13-32 ] asymptotic properties of a simultaneous e
stimator $(\\lambda_n\;\\beta_n)$ for the baseline hazard rate $\\lambda
(\\cdot)$ and the regression parameter $\\beta$ were studied\, at that the
parameter set $\\Theta=\\Theta_{\\lambda}\\times \\Theta_{\\beta}$ was as
sumed bounded.\n\nIn Kukush and Chernova (2017) [ Theory of Probability an
d Mathematical Statistics **96**\, 100-109 ] we dealt with the simultaneo
us estimator $(\\lambda_n\;\\beta_n)$ in the case\, where the $\\Theta_{\
\lambda}$ was unbounded from above and not separated away from $0$. The e
stimator was constructed in two steps: first we derived a strongly consist
ent estimator and then modified it to provide its asymptotic normality. \n
\nIn this talk\, we construct the confidence interval for an integral fun
ctional of $\\lambda(\\cdot)$ and the confidence region for $\\beta$. We
reach our goal in each of the three cases: (a) the measurement error is
bounded\, (b) it is normally distributed\, or (c) it is a shifted Poisson
random variable. The censor is assumed to have a continuous pdf. In future
research we intend to elaborate a method for heavy tailed error distribu
tions.\n\nhttps://indico.uu.se/event/317/contributions/297/
LOCATION:Ångströmslaboratoriet
URL:https://indico.uu.se/event/317/contributions/297/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Information criteria for structured sparse variable selection
DTSTART;VALUE=DATE-TIME:20170817T123000Z
DTEND;VALUE=DATE-TIME:20170817T130000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-301@indico.uu.se
DESCRIPTION:Speakers: Bastien Marquis (Université Libre de Bruxelles)\, M
aarten Jansen (Université Libre de Bruxelles)\n\nIn contrast to the low d
imensional case\, variable selection under the \nassumption of sparsity in
high dimensional models is strongly influenced by the\neffects of false p
ositives.\nThe effects of false positives are tempered by combining the va
riable selection\nwith a shrinkage estimator\, such as in the lasso\, wher
e the selection is\nrealized by minimizing the sum of squared residuals re
gularized by an $\\ell_1$\nnorm of the selected variables. Optimal variabl
e selection is then equivalent\nto finding the best balance between closen
ess of fit and regularity\, i.e.\, to\noptimization of the regularization
parameter with respect to an information\ncriterion such as Mallows's Cp o
r AIC. For use in this optimization\nprocedure\, the lasso regularization
is found to be too tolerant towards false\npositives\, leading to a consid
erable overestimation of the model size. Using an\n$\\ell_0$ regularizatio
n instead requires careful consideration of the false\npositives\, as they
have a major impact on the optimal regularization parameter.\nAs the fram
ework of the classical linear model has been analysed in previous\nwork\,
the current paper concentrates on structured models and\, more \nspecifica
lly\, on grouped variables. Although the imposed structure in the \nselect
ed models can be understood to somehow reduce the effect of false\npositiv
es\, we observe a qualitatively similar behavior as in the unstructured\nl
inear model.\n\nhttps://indico.uu.se/event/317/contributions/301/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/301/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Inference on covariance matrices and operators using concentration
inequalities
DTSTART;VALUE=DATE-TIME:20170816T070000Z
DTEND;VALUE=DATE-TIME:20170816T073000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-302@indico.uu.se
DESCRIPTION:Speakers: Adam Kashlak (Cambridge Centre for Analysis\, Univer
sity of Cambridge)\n\nIn the modern era of high and infinite dimensional d
ata\, classical statistical\nmethodology is often rendered inefficient and
ineffective when confronted\nwith such big data problems as arise in geno
mics\, medical imaging\, speech\nanalysis\, and many other areas of resear
ch. Many problems manifest when\nthe practitioner is required to take into
account the covariance structure\nof the data during his or her analysis\
, which takes on the form of either a\nhigh dimensional low rank matrix or
a finite dimensional representation of\nan infinite dimensional operator
acting on some underlying function space.\nThus\, we propose using tools f
rom the concentration of measure literature\nto construct rigorous descrip
tive and inferential statistical methodology for\ncovariance matrices and
operators. A variety of concentration inequalities are\nconsidered\, which
allow for the construction of nonasymptotic dimension-free\nconfidence se
ts for the unknown matrices and operators. Given such confidence\nsets a w
ide range of estimation and inferential procedures can be and are\nsubsequ
ently developed.\n\nhttps://indico.uu.se/event/317/contributions/302/
LOCATION:Museum Gustavianum (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/302/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Methods for bandwidth detection in kernel conditional density esti
mations
DTSTART;VALUE=DATE-TIME:20170815T113000Z
DTEND;VALUE=DATE-TIME:20170815T120000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-303@indico.uu.se
DESCRIPTION:Speakers: Katerina Konecna (Masaryk University)\n\nThis contri
bution is focused on the kernel conditional density estimations (KCDE). Th
e estimation depends on the smoothing parameters which influence the final
density estimation significantly. This is the reason why a requirement of
any data-driven method is needed for bandwidth estimation. In this contri
bution\, the cross-validation method\, the iterative method and the maximu
m likelihood approach are conducted for bandwidth selection of the estimat
or. An application on a real data set is included and the proposed methods
are compared.\n\nhttps://indico.uu.se/event/317/contributions/303/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/303/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Invited Speaker - Embedding machine learning in stochastic process
algebra
DTSTART;VALUE=DATE-TIME:20170816T110000Z
DTEND;VALUE=DATE-TIME:20170816T120000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-304@indico.uu.se
DESCRIPTION:Speakers: Jane Hillston (University of Edinburgh)\n\nhttps://i
ndico.uu.se/event/317/contributions/304/
LOCATION:Museum Gustavianum
URL:https://indico.uu.se/event/317/contributions/304/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Invited speaker - Formal languages for stochastic modelling
DTSTART;VALUE=DATE-TIME:20170816T090000Z
DTEND;VALUE=DATE-TIME:20170816T100000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-305@indico.uu.se
DESCRIPTION:Speakers: Jane Hillston (University of Edinburgh)\n\nhttps://i
ndico.uu.se/event/317/contributions/305/
LOCATION:Museum Gustavianum
URL:https://indico.uu.se/event/317/contributions/305/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Invited Speaker - Non-limiting spatial extremes
DTSTART;VALUE=DATE-TIME:20170815T070000Z
DTEND;VALUE=DATE-TIME:20170815T080000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-306@indico.uu.se
DESCRIPTION:Speakers: Jenny Wadsworth (Lancaster University)\n\nMany quest
ions concerning environmental risk can be phrased as spatial extreme value
problems. Classical extreme value theory provides limiting models for max
ima or threshold exceedances of a wide class of underlying spatial process
es. These models can then be fitted to suitably defined extremes of spatia
l datasets and used\, for example\, to estimate the probability of events
more extreme than we have observed to date. However\, a major practical pr
oblem is that frequently the data do not appear to follow these limiting m
odels at observable levels\, and assuming otherwise leads to bias in estim
ation of rare event probabilities. To deal with this we require models tha
t allow flexibility in both what the limit should be\, and in the mode of
convergence towards it. I will present a construction for such a model and
discuss its application to some wave height data from the North Sea.\n\nh
ttps://indico.uu.se/event/317/contributions/306/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/306/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Invited speaker - Sequential Monte Carlo: basic principles and alg
orithmic inference
DTSTART;VALUE=DATE-TIME:20170814T110000Z
DTEND;VALUE=DATE-TIME:20170814T120000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-307@indico.uu.se
DESCRIPTION:Speakers: Jimmy Olsson (Royal Institute of Technology)\n\nSequ
ential Monte Carlo methods form a class of genetic-type algorithms samplin
g\, on-the-fly and in a very general context\, sequences of probability me
asures. Today these methods constitute a standard device in the statistici
an's tool box and are successfully\n applied within a wide range of scient
ific and engineering disciplines. This talk is split into two parts\, wher
e the first provides an introduction to the SMC methodology and the second
discusses some novel results concerning the stochastic stability and vari
ance\n estimation in SMC.\n\nhttps://indico.uu.se/event/317/contributions/
307/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/307/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Invited speaker - Sequential Monte Carlo: basic principles and alg
orithmic inference
DTSTART;VALUE=DATE-TIME:20170814T090000Z
DTEND;VALUE=DATE-TIME:20170814T100000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-308@indico.uu.se
DESCRIPTION:Speakers: Jimmy Olsson (Royal Institute of Technology)\n\nSequ
ential Monte Carlo methods form a class of genetic-type algorithms samplin
g\, on-the-fly and in a very general context\, sequences of probability me
asures. Today these methods constitute a standard device in the statistici
an's tool box and are successfully\n applied within a wide range of scient
ific and engineering disciplines. This talk is split into two parts\, wher
e the first provides an introduction to the SMC methodology and the second
discusses some novel results concerning the stochastic stability and vari
ance\n estimation in SMC.\n\nhttps://indico.uu.se/event/317/contributions/
308/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/308/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Delete or Merge Regressors algorithm
DTSTART;VALUE=DATE-TIME:20170815T083000Z
DTEND;VALUE=DATE-TIME:20170815T090000Z
DTSTAMP;VALUE=DATE-TIME:20221207T142300Z
UID:indico-contribution-309@indico.uu.se
DESCRIPTION:Speakers: Agnieszka Prochenka (Warsaw University)\n\nThis pape
r addresses a problem of linear and logistic model selection in the presen
ce of both continuous and categorical predictors. In the literature two ty
pes of algorithms dealing with this problem can be found. The first one we
ll known group lasso (\\cite{group}) selects a subset of continuous and a
subset of categorical predictors. Hence\, it either deletes or not an ent
ire factor. The second one is CAS-ANOVA (\\cite{cas}) which selects a subs
et of continuous predictors and partitions of factors. Therefore\, it merg
es levels within factors. Both these algorithms are based on the lasso reg
ularization.\n\nIn the article a new algorithm called DMR (Delete or Merge
Regressors) is described. Like CAS-ANOVA it selects a subset of continuou
s predictors and partitions of factors. However\, instead of using regular
ization\, it is based on a stepwise procedure\, where in each step either
one continuous variable is deleted or two levels of a factor are merged. T
he order of accepting consecutive hypotheses is based on sorting t-statist
ics or linear regression and likelihood ratio test statistics for logistic
regression. The final model is chosen according to information criterion.
Some of the preliminary results for DMR are described in \\cite{pro}.\n\n
DMR algorithm works only for data sets where $p < n$ (number of columns in
the model matrix is smaller than the number of observations). In the pape
r a modification of DMR called DMRnet is introduced that works also for da
ta sets where $p \\gg n$. DMRnet uses regularization in the screening step
and DMR after decreasing the model matrix to $p < n$.\n\nTheoretical resu
lts are proofs that DMR for linear and logistic regression are consistent
model selection methods even when $p$ tends to infinity with $n$. Furtherm
ore\, upper bounds on the error of selection are given.\n\nPractical resul
ts are based on an analysis of real data sets and simulation setups. It is
shown that DMRnet chooses smaller models with not higher prediction error
than the competitive methods.\nFurthermore\, in simulations it gives most
often the highest rate of true model selection.\n\nhttps://indico.uu.se/e
vent/317/contributions/309/
LOCATION:Ångströmslaboratoriet (Uppsala University)
URL:https://indico.uu.se/event/317/contributions/309/
END:VEVENT
END:VCALENDAR