|
Thursday, October 22nd 2026
14:00-16:15
|
|
Luís Meira-Machado Full Professor at DMAT and Researcher at CMAT - University of Minho
This course provides a practical introduction to the use of artificial intelligence tools in statistical
learning, research, and professional practice. Participants will explore how AI can support literature reviews, scientific writing, statistical programming, and data
analysis. Through a hands-on example in survival analysis, they will use ChatGPT and Codex to develop, verify, and improve R code, and conduct simulation studies in
RStudio. Particular emphasis will be placed on the limitations of AI, the importance of critical evaluation, and the need to validate AI-generated results.
|
16:50
|
|
CMAT Research Highlights
This session features a series of short presentations by researchers from CMAT and the University of Minho.
It provides an opportunity to share ongoing research, recent results, and new ideas across different areas of Mathematics and Statistics.
|
Friday, October 23rd 2026
09:15-09:45
|
|
Luís Meira-Machado - President of the Portuguese Statistical Society
Abstract: The opening session of the Workshop on Statistics and Data Science 2026 will be led by the
President of the Portuguese Statistical Society. In addition to offering a warm welcome to all participants and officially launching the event, the session will
highlight the significance of the European Statistics Day, celebrated on October 20th. This day underscores the vital role of statistics in society, aligning with the
workshop's focus on advancing knowledge in Statistics, Applied Probability, and Operational Research.
|
09:45-10:25
|
|
Pércio António Chitata - Universidade Púnguè, Moçambique
Abstract: A expansão do ensino superior em Moçambique nas últimas décadas tem sido acompanhada por uma
crescente complexidade institucional, caracterizada pelo aumento do número de instituições, estudantes, cursos, docentes, áreas científicas e unidades de apoio à missão
universitária. Esta evolução exige a conceção de sistemas estatísticos capazes de produzir informação fiável, consistente, comparável e útil para diferentes níveis de
decisão, desde a gestão interna das instituições até à formulação de políticas públicas nacionais. As instituições de ensino superior (IES) produzem diariamente grandes
volumes de dados administrativos e académicos. Contudo, o desafio atual não reside apenas na recolha desses dados, mas na capacidade de transformá-los em informação
estatística de qualidade, capaz de apoiar a planificação estratégica, a monitoria de metas institucionais, a avaliação da qualidade, a prestação de contas, a produção de
estatísticas nacionais e a comparação internacional. Esta comunicação analisa os principais desafios associados à produção e utilização de estatísticas no ensino superior
moçambicano, tomando como referência a experiência da Universidade Púnguè. Discute a arquitetura atual de produção estatística, os modelos de circulação da informação entre
as IES e os órgãos nacionais responsáveis pela produção estatística, bem como os desafios relacionados com a interoperabilidade dos sistemas de informação académica, a
padronização das classificações educativas, a qualidade dos dados e a construção de indicadores institucionais. A partir do caso da Universidade Púnguè, a comunicação
apresenta desafios concretos relacionados com a integração de diferentes fontes de informação, limitações dos sistemas existentes, necessidade de desenvolvimento de
capacidades técnicas e criação de mecanismos complementares de produção estatística. Igualmente, discute oportunidades para a aplicação de métodos estatísticos, ciência
de dados e investigação operacional na gestão universitária, incluindo técnicas de análise exploratória, modelação estatística, análise multivariada, previsão e otimização.
Adicionalmente, a comunicação procura contribuir para o debate sobre o fortalecimento do ecossistema estatístico das instituições de ensino superior em contextos
emergentes, parindo da colaboração entre instituições académicas, unidades de investigação e centros especializados. Palavras-chave: Estatística aplicada; ensino
superior; qualidade dos dados; indicadores institucionais; ciência de dados; Moçambique.
|
11:05-11:45
|
|
Juan Carlos Pardo-Fernández, Department of Statistics and OR - University of Vigo, Vigo, Spain
Abstract: The Receiver Operating Characteristic (ROC) curve is a widely used statistical tool for assessing the performance of a binary classification procedure
based on a continuous marker. In the last decades, functional markers have become a frequent tool in medical diagnosis. In this talk, we aim to define a discriminating
index to distinguish two populations when the observations are functional data. We discuss some of the problems arising when estimating optimal directions defined to
maximize the AUC of a projection index and we construct the corresponding ROC curve. Under mild conditions, consistency results for the ROC estimators are derived for
general indexes, including both linear and quadratic ones. Numerical experiments allow to see the advantages of the quadratic rule when the populations have different
covariance operators. We also illustrate the considered methods on a real data set on cardiotoxicity related to therapies on breast cancer patients. This is joint work
with Ana M. Bianco and Graciela Boente, Universidad de Buenos Aires.
|
11:45-12:10
|
|
Carolina S. Marques - Centro de Estatística e Aplicações, Universidade de Lisboa, Lisboa, Portugal; Instituto Universitário de Lisboa (Iscte-IUL), Departamento de Métodos Quantitativos,
Lisboa, Portugal
Abstract: The study of dinosaurs often relies on scarce and fragmented fossil records, posing significant
challenges for fossil classification and the identification of new fossil sites. Machine learning and statistical modelling offer promising approaches to address these
challenges. This talk presents applications of convolutional neural networks and random forests for classifying dinosaur teeth and footprints, alongside Bayesian spatial
models designed to identify areas with a higher probability of new fossil discoveries.
|
14:15-14:55
|
|
Tiago Dias Domingues - CEAUL, Faculdade de Ciências, Universidade de Lisboa, Portugal
Abstract: Serological tests measure antibody levels that rarely follow symmetric distributions, which undermines classical approaches to separating seronegative from
seropositive individuals, such as the "mean plus three standard deviations" rule. In this seminar, I present finite mixture models based on Skew-Normal and Skew-t
distributions, fitted via the ECM algorithm. Three methods are compared for estimating the cutoff point when the true serological status is unknown: the 99.9% quantile
of the seronegative population, the minimum of the mixture density, and conditional classification curves. Their performance is assessed through a Monte Carlo simulation
study across different sample sizes and seropositive proportions. The methods are then applied to IgG antibody data against four SARS-CoV-2 antigens, where the true
infection status is known, and compared with the ROC curve approach. Implications for seroprevalence estimation are discussed.
|
14:55-15:35
|
|
Alba Fuster Alonso - Instituto de Ciencias del Mar (ICM) - CSIC, Barcelona, Spain
Abstract: Species distribution models (SDMs) are key for understanding and predicting species' spatio-temporal patterns, especially under global change. Although SDMs
have been widely applied on regional scales, there is a growing need to extend these models globally to capture the large-scale environmental and ecological processes
driving marine species distributions. To move toward more realistic global marine SDMs, we need methodological improvements, such as accounting for Earth's spherical
geometry and considering dispersal constraints due to geographic barriers. We present a novel approach to global species distribution modeling by integrating
spatio-temporal barriers into the INLA-SPDE (Integrated Nested Laplace Approximation - Stochastic Partial Differential Equation) framework. This method allows for
efficient Bayesian inference of spatial and temporal dynamics while explicitly incorporating physical barriers and Earth's spherical geometry. By building a worldwide
mesh on spherical coordinates and anisotropic covariance structures, we ensure biologically realistic connectivity across global ocean basins. We assess the
capabilities of our model through a comprehensive simulation study, evaluating its performance under both continuous and binary response scenarios. Furthermore,
we apply our method to two case studies, Atlantic cod (Gadus morhua) and shallow coral species, which demonstrate improved spatial predictions and ecological realism
compared to models that ignore barriers and Earth's spherical geometry. Our findings emphasize the importance of incorporating geographical and ecological constraints
in global SDMs, providing a computationally efficient framework for ecological.
|
16:15-16:40
|
|
Patrícia Neto - PhD program in Mathematics, Centre of Mathematics, Department of Mathematics, University of Minho, Portugal
Abstract: Delay-coordinate representations combine current and past observations, but selecting a delay from an estimated
dependence curve can be sensitive to finite-sample effects. This preliminary study examines mutual-information estimation and the first-local-minimum rule using four stochastic
benchmark models with known population dependence. Histogram, kernel-density and nearest-neighbour estimators are considered, distinguishing estimation accuracy from delay-selection
reliability. Weekly e-commerce product data provide the application context, with an illustrative example showing how current and past observations are combined into uniform delay
vectors.
|
16:40-17:05
|
|
Elsa Soares - PhD in Applied Mathematics, Centre of Mathematics, Department of Mathematics, University of Minho, Portugal
Abstract: Longitudinal data are characterized by repeated measurements over time. Traditional statistical models applied to longitudinal data have been widely used,
but are notable limitations. Machine learning techniques are emerging as promising alternatives, especially for high-dimensional data. This study reviews these
methodologies, comparing them with traditional models, and discusses challenges such as correlation among repeated measures and missing data, highlighting their
clinical relevance.
|
|