Ciclo de Palestras – Primeiro Semestre de 2008

As palestras são realizadas na sala C-116 do Centro Tecnológico as 15:30 h, a menos que ocorra aviso em contrário.

In this presentation I shall review the methods proposed in the literature for small area estimation. The following approaches and methods will be covered: 1. Design-based methods, advantages and limitations 2. Frequentist model based methods 3. Bayesian model-based methods 4. Methods using time series/cross-sectional models 5. Recent advances: Accounting for measurement errors and outliers, choice of priors, ranking, M-quantiles,…
O Painel Intergovernamental de Mudanças Climáticas (IPCC) aponta, claramente, para cenários climáticos para os próximos 100 anos nos quais o efeito antrópico é evidente. Os cenários do IPCC são mais sólidos no que se refere às características médias do clima terrestre. Por exemplo, a temperatura media da terra deve aumentar assim como as chuvas. Entretanto, ninguém vive no chamado “clima médio terrestre” e existe incerteza significativa quanto aos efeitos regionais apesar de ter havido grande no último relatório quanto aos aspectos climáticos regionais. O IPCC também aponta para extremos climáticos mais freqüentes (secas , enchentes, ondas de calor e mesmo de frio). Portanto, é evidente que é preciso estabelecer políticas públicas que nos preparem para situações climáticas adversas. Em primeiro lugar, devemos aprender a conviver com a variabilidade climática natural. É preciso reconhecer que já tivemos flutuações climáticas de causas estritamente naturais de grande porte. Por ex., a década de 40-60 foi cerca de 15% mais seca no SE brasileiro que o período 70-90. Houve evidente impacto na utilização de recursos hídricos para a geração de energia elétrica e na agricultura. Esses eventos do passado recente precisam ser melhor estudados, do ponto de vista dos impactos sociais e econômicos. Outro aspecto fundamental que as políticas públicas devem abordar: quanto investir na redução das emissões e na adaptação? O custo da descarbonização da nossa economia precisa ser melhor avaliado. O custo da adaptação também necessita melhor quantificação. A decisão quanto ao caminho futuro depende de uma análise critica desses custos e das incertezas nos cenários futuros.

Near-critical percolation and the geometry of diffusion fronts
Pierre Nolin (ENS-Paris)
We discuss a model of inhomogeneous medium known as “Gradient Percolation”, which is an inhomogeneous percolation process where the den- sity of occupied sites depends on the location in space. This model was ¯rst introduced by the physicists Gouyet, Rosso and Sapoval in 1985 to show nu- merical evidence that diffusion fronts are fractal. The macroscopic interface – separating occupied sites and vacant sites – that appears remains localized in regions where the density of occupied sites is close to the percolation threshold pc, and its behavior can be described using properties of near-critical standard percolation. This allows to study a simple two-dimensional model where a large number of particles that start at a given site diffuse independently. As the particles evolve, a concentration gradient appears and we observe a macroscopic interface. We exhibit in particular a regime where this (properly rescaled) interface is fractal with dimension 7/4: this model thus provides a natural setting where fractal geometry spontaneously arises, as predicted by physicists.

Are random frontiers always symmetric?
Wendelin Werner, Medalha Fields 2006 (ENS-Paris)
We will make comments on the following problem: Suppose that a two- dimensional domain is created at random from many infinitesimal local inputs, and that one only observes its outer boundary. Can one detect on which side the domain is by looking at a portion of this boundary?
The talk is partly based on joint work with Pierre Nolin.

The potentially large economic impacts of animal disease epidemics have been highlighted in recent years through outbreaks such as foot-and-mouth disease (FMD) in the UK during 2001. This talk reports work from a project with the Veterinary Laboratories Agency (VLA), Weybridge, UK which was concerned with use of survival modelling to develop dynamic space-time predictions of survivor and hazard functions for individual farm premises as an FMD epidemic progresses. Such survival analyses could provide powerful insights into the patterns of infection, and assist in optimising various aspects of the operational response activities, such as targeting of `at-risk’ premises. The talk will discuss and compare various possible model formulations using both simulated and real data based on the 2001 UK FMD epidemic and then go on to illustrate how model predictions may be used to refine response activities.
The talk considers the general model of reduced rank regression (RRR). RRR is useful to model the asymmetric linear relationship between two sets of predictants and predictors. Sometimes there is a need to reduce the number of parameters in the multivariate regression model imposing constraints on the rank, r, of the matrix of coefficients. A special case of interest is when r=1, which allows for the discussion of a model with applications in multi-output stochastic frontier analysis and spatial regression models.