[ABE-L] Sessão de Estatística 09/05 - Celebrando Mulheres na Matemática - IM-UFRJ

[Alexandra M. Schmidt]

Prezada(o)s, Colegas,

Entre os dias 09 e 13 de maio acontecerá na UFRJ o Workshop  Celebrando
Mulheres na Matemática <http://www.dinamicas.im.ufrj.br/celebra-cwinm/> .

O evento será* online* e as inscrições podem ser feitas aqui
<https://forms.gle/YX1wJg64uVTv6Tw89>.

Na segunda-feira, 09/05, entre 10:15h e 12:15h  teremos a Sessão de
Estatística do evento, onde teremos as seguintes palestras *(resumos seguem
após final da mensagem)*:



   - *Mariane B. Alves (IM-UFRJ, Brasil)*
   - k-parametric dynamic generalized linear models: a sequential approach
   via information geometry


   - *Nancy L. Garcia (IMECC, Unicamp, Brasil)*
   - Bayesian analysis of Brazilian and European written texts


   - *Camila P.E. de Souza (University of Western Ontario, Canadá)*
   - Clustering functional data via variational inference
   -
   - *Lelys Bravo de Guenni (University of Illinois at Urbana-Champaign,
   EUA)*
   - Deep Lagged-Wavelet for monthly rainfall forecasting in a tropical
   region


Contamos com sua presença!

Abraços
Alexandra
Alexandra M. Schmidt
Professor - The University Chair
Program Director of Biostatistics
McGill University

*http://alex-schmidt.research.mcgill.ca/
<http://alex-schmidt.research.mcgill.ca/>*
*[image: image.jpeg]*
*https://isbawebmaster.github.io/ISBA2022/
<https://isbawebmaster.github.io/ISBA2022/>*

*Mariane B. Alves (IM-UFRJ)*

*k-parametric dynamic generalized linear models: a sequential approach via
information geometry*

Dynamic generalized linear models may be seen as an extension to dynamic
linear models, accommodating non-Gaussian responses, and to generalized
linear models, formally treating serial auto-correlation inherent to
responses in the exponential family, observed through time. The Bayesian
inference scheme does not have analytical solution and there are several
numerical approximating proposals in the literature, many of them relying
on Monte Carlo Markov Chain, with the burden of computational cost. In this
talk, a new approach based on information geometry, focusing on the
k-parametric exponential family, will be presented. Among others, the
proposed method accommodates multinomial responses on k=d+1 categories and
normal responses with dynamic predictive structure for the mean as well as
for the precision parameter. The method preserves the sequential aspect of
the Bayesian inferential procedure, producing real-time inference and
naturally accommodates association among k predictors. Information geometry
concepts such as Kullback-Leibler divergence and the projection theorem are
used in the development of the method, placing it close to recent
approaches of variational inference. The method is computationally
efficient and flexible to quickly accommodate new patterns and information
when strategically needed, favorably comparing to alternative approaches in
the literature, preserving aspects of monitoring and intervention analysis,
which are usual in sequential analyzes. This is joint work with Raı́ra
Marotta and Helio dos Santos Migon.


*Nancy L. Garcia (IMECC, Unicamp, Brazil)*

*Bayesian analysis of Brazilian and European written texts*

Are Brazilian and European Modern Portuguese different? Can we identify
these differences through rhythmic characteristics of written texts? We
answer these questions by modelling some characteristics that can be
extracted automatically from written texts as a Markov Chain of Variable
Length. Markov chains with variable length are useful parsimonious
stochastic models able to generate most stationary sequence of discrete
symbols. The idea is to identify the suffixes of the past, called contexts,
that are relevant to predict the future symbol. This is joint work with
Victor Freguglia.


*Camila P.E. de Souza (University of Western Ontario)*

*Clustering functional data via variational inference*

Functional data analysis (FDA) deals with data recorded densely over time
(or any other continuum) with one or more observed curves per subject.
Conceptually, functional data are continuously defined, but in practice,
they are usually observed at discrete points. Among different kinds of
functional data analyses, clustering analysis aims to determine underlying
groups of curves in the data when there is no information on the group
membership of each individual curve. In this work, we propose a new
model-based approach for clustering and smoothing functional data
simultaneously via variational inference (VI). Therefore, we derive a
coordinate ascent variational inference (CAVI) algorithm to approximate the
posterior distribution of our model parameters by finding the variational
distribution with the smallest Kullback-Leibler divergence to the
posterior. To our best knowledge, there are no studies in the literature on
clustering functional data through VI. Our CAVI algorithm is implemented as
an R package, and its performance is evaluated using simulated data and
publicly available datasets. This is joint work with my students Chengqian
Xian and John T. Jewell, and collaborators Ronaldo Dias and Adriano Zambom.


*Lelys Bravo de Guenni (University of Illinois at Urbana-Champaign, USA)*

Deep Lagged-Wavelet for monthly rainfall forecasting in a tropical region
Rainfall forecasting is an important decision-making input in a variety of
areas, including agriculture, hydropower generation, and water resource
planning and management. A reliable forecasting tool would contribute to
the reduction of vulnerability and risk in water management systems.
However, due to the high spatial-temporal variability of rainfall amounts,
it is very difficult to achieve high accuracy in the forecasts. This study
addresses the problem of rainfall forecasting by proposing a methodology
based on a combination of wavelet decomposition (WD), neural networks (NN),
and lagged regression (LR). We implemented WD in a pre-processing phase
followed by the use of a recurrent NN algorithm, and proposed a prediction
enhancement phase by optimizing the network outputs using a monthly
rainfall forecast correction with LR. The methodology was implemented at
four weather stations in a tropical region, and it was compared with other
powerful forecasting methods. The research results suggest that our
approach outperformed other methods in performance accuracy and biases
correction, achieving adjusted R squared greater than 0.76 and normalized
mean absolute errors less than 0.31.
-------------- Pr�xima Parte ----------
Um anexo em HTML foi limpo...
URL: <http://lists.ime.usp.br/pipermail/abe/attachments/20220430/50f016ce/attachment-0001.htm>
-------------- Pr�xima Parte ----------
Um anexo não-texto foi limpo...
Nome: image.jpeg
Tipo: image/jpeg
Tamanho: 59456 bytes
Descrição: não disponível
URL: <http://lists.ime.usp.br/pipermail/abe/attachments/20220430/50f016ce/attachment-0001.jpeg>