Department of Statistics and Operations Research

Seminar Schedule: Year 2009


MITIGATION OF HAZARDOUS MATERIALS TRANSPORT RISK VIA ROAD NETWORK DESIGN

CONVIDAT: Vedat Verter, Desautels Faculty of Management, McGill University, Canada

IDIOMA: Anglès.

LLOC: Edifici C5, Aula: C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 18 de desembre de 2009, Hora: 13:00

RESUM: Hazardous materials (hazmats) are an integral part of our industrialized lifestyle and large quantities of them need to be transported between the points of their production to the points of their consumption or disposal. Although the transportation sector is mostly deregulated, hazmat shipments remain regulated by government agencies in many countries, mainly due to the associated public and environmental risks. One of the policy tools that are available to a government agency is to close certain road segments to the shipments that involve certain types of hazmats. The presentation will focus on three alternative approaches for modeling and solving the regulators? problem of identifying the most appropriate road segments to apply such curfews while maintaining the economic viability of the sector and minimizing the population exposure to hazmat transport risks. These approaches involve the use of (i) a bilevel programming model, (ii) a path-based formulation, and (iii) a tariff-based model for hazmat network design. Our findings through the implementation of these alternative approaches on the highway networks of the provinces of Quebec and Ontario in Canada will also be discussed.

Web del ponent: clica aquí

Inici
MUESTREO ESPACIAL ASISTIDO Y BASADO EN MODELOS

CONVIDAT: Luís Ambrosio, Departamento de Economía y Ciencias Sociales Agrarias, Universidad Politécnica de Madrid.

IDIOMA: Castellà

LLOC: FME, Edifici U, Aula: 004, Campus Sud, UPC, Pau Gargallo, 5, 02028 Barcelona (veure mapa)

DATA: Dimecres, 16 de desembre de 2009, Hora: 12:30

RESUM: Para la estimación por muestreo de características que se distribuyen en el espacio (muestreo espacial) se sugiere una aproximación asistida por modelos, si la población y la muestra son grandes y una aproximación basada en modelos, para la estimación en áreas pequeñas o dominios de estudio. Ambas aproximaciones están bien establecidas para variables continuas, pero no para variables discretas (variables de conteo y cualitativas). Se propone extender a las variables discretas los resultados establecidos para las variables continuas. En particular, se considera una aproximación basada en el diseño de muestras sistemáticas asistido por modelos de Poisson con correlación espacial, para la estimación de variables de conteo en poblaciones espaciales, y una aproximación basada en modelos binomiales y multinomiales con correlación espaciotemporal, para la estimación de proporciones con muestras de panel georeferenciadas.

Inici
SAT Y PROGRAMACIÓN CON RESTRICCIONES: 2 ALTERNATIVAS A LA PROGRAMACIÓN ENTERA

CONVIDAT: Javier Larrosa, Dept. of Languages and Computer Systems, Universitat Politècnica de Catalunya.

IDIOMA: Castellà

LLOC: Edifici C5, Aula: C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 27 de novembre de 2009, Hora: 12:00

RESUM: SAT-solving y Constrant Programming (CP) son dos tecnologías que están emergiendo con fuerza como alternativa o complemento a la Programación Entera. El origen de SAT está en la lógica computacional y el de CP en los lenguajes de programación y la Inteligencia Artificial. En esta charla haré una breve introducción de ambas tecnologías y daré mi opinión sobre sus ventajas e inconvenientes. El objetivo de la charla no es tanto entrar en detalles técnicos sino entender las ideas e hipótesis subyacentes a cada aproximación.

Web del ponent: clica aquí

Inici
THE OPEN VEHICLE ROUTING PROBLEM AND THE DISRUPTED VEHICLE ROUTING PROBLEM

CONVIDAT: Richard Eglese, Dept. of Management Science, Lancaster University Management School, UK

IDIOMA: Anglès

LLOC: Edifici C5, Aula: C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 23 d'Octubre de 2009, Hora: 12:00

RESUM: When distributing goods from a central depot, in the Open Vehicle Routing Problem vehicles are not required to return to the depot after completing deliveries to customers. Heuristic and exact algorithms will be presented for this problem and compared with the closed version of the problem. Another type of vehicle routing problem is discussed: the Disrupted Vehicle Routing Problem where there may be various reasons for a planned set of vehicle trips to be disrupted. In particular, a disruption where a vehicle breaks down while on the road and a new routing solution needs to be quickly generated will be examined. A heuristic algorithm based on tabu search has been developed to solve the problem. A set of test problems has been generated and computational results from experiments using the heuristic algorithm are presented. The connection between these two problems will be discussed.

Web del ponent: clica aquí

TRANSPARENCIES DE LA PRESENTACIO: clica aquí

Inici
TRANSFORMATION KERNEL ESTIMATION OF ACTUARIAL LOSS FUNCTIONS; PDF AND CDF KERNEL ESTIMATION

CONVIDAT: Catalina Bolancé

LLOC: Edifici C5, Aula: C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 16 d'Octubre de 2009, Hora: 12:30

RESUM: A Transformation kernel PDF and CDF estimator that is suitable for heavy-tailed distributions is presented. Alternatively to other transformations, we propose using a double transformation, the choice of the bandwidth parameter becomes straightforward. Some illustrations and simulation results are presented.

Inici
SHOULD THE ANALYSES OF MULTI-CENTER RANDOMIZED CLINICAL TRIALS BE GUIDED BY THE STUDY DESIGN? BASIC IDEAS AND INSIGHTS

CONVIDAT: Marvin Zelen

IDIOMA: Anglès

LLOC: Edifici C5, Aula: C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 9 d'Octubre de 2009, Hora: 12:30

RESUM: Most investigators would agree that the analysis of randomized multi -center clinical trials should be guided by the design of the trial. Many trials are designed using permuted blocks in which the allocation is over time. Yet this feature is ignored in the analysis. The randomization process is also ignored in most analyses. Also there is general agreement that analyses should account for the factors which are important in affecting the principal endpoints. Current practice is to ignore the effect of institutional variation. The institutional effects are ignored because many trials may have large numbers of centers which enter a relatively small number of patients making it difficult to estimate the magnitude of the institutional variation. Yet we know in drug trials patient support and management may widely differ among centers resulting in potential institutional variation. All of the statistical analyses methods require a random sample of patients. However it is rare for a trial to have a random sample of patients. Consequently an issue arises as to whether the trial conclusions apply to the eligible population with disease or simply to the patients who have been entered in the trial. New methods will be discussed which rely on randomization as the basis of inference and at the same time take into account the design of the trial

Web del ponent: clica aquí per accedir a la web de Marvin Zelen.

Inici
INNOVATIONS IN CURRICULUM DESIGN: CONTEXTUALIZED LEARNING AND THE USE OF AN ONLINE LEARNING ENVIRONMENT

CONVIDAT: Jenny Freeman

IDIOMA: Anglès

LLOC: FME, Edifici U, Aula: 005, Campus Sud, UPC, Pau Gargallo, 5, 02028 Barcelona (veure mapa)

DATA: Dimecres, 15 de Juliol de 2009, Hora: 12:30

RESUM: There can be little doubt that statistics is important and relevant to many college subjects. Coincidently, it is increasingly common for students to dislike and under-perform in modules involving mathematics, numeracy or statistics, and it is increasingly clear that teaching statistics in a traditional didactic way does not entirely meet their needs; many fail to understand basic concepts, which impacts negatively on their use and interpretation of statistics. This talk will describe a recent project to redesign the way in which statistics is taught to students in disciplines other than statistics, primarily medical undergraduates, in order to make the subject more relevant to them. The project brought together a statistician, a physician and experts from the Learning Development and Media Unit at the University of Sheffield to produce a new learning package for medical students. The project explored ways of enhancing the teaching and learning of statistics to enable students to better apply their statistical knowledge. It re-conceptualized the syllabus by focusing on developing different methods of delivery, including using a web-based learning environment, and allowed students to access materials at levels consistent with their current understanding. Development of new and remodelled teaching materials, and approaches placed greater emphasis on the application of statistics and interpretation of data rather than the mechanics. As part of the teaching, videos and animations were developed to show how statistics are used during consultations and when making medical decisions and in addition the teaching is delivered by both a statistician and clinician who together explain the statistical concepts and their relevance to medical practice. And finally an interactive online resource was created for the students to access outside of the taught sessions.

Web del ponent: clica aquí per accedir a la web de la Jenny Freeman.

Inici
GENERALIZED SPATIAL DYNAMIC FACTOR MODELS

CONVIDAT: Hedibert Freitas Lopes

IDIOMA: Anglès

LLOC: FME, Edifici U, Aula: S01, Campus Sud, UPC (veure mapa)

DATA: Divendres, 26 de Juny de 2009, Hora: 13:00

RESUM: This paper introduces a new class of spatio-temporal models for measurements belonging to the exponential family of distributions. In this new class, the spatial and temporal components are conditionally independently modeled via a latent factor analysis structure for the (canonical) transformation of the measurements mean function. The factor loadings matrix is responsible for modeling spatial variation, while the common factors are responsible for modeling the temporal variation. One of the main advantages of our model with spatially structured loadings is the possibility of detecting similar regions associated to distinct dynamic factors. We also show that the new class outperforms a large class of spatial-temporal models that are commonly used in the literature. Posterior inference for fixed parameters and dynamic latent factors is performed via a custom tailored Markov chain Monte Carlo scheme for multivariate dynamic systems that combines extended Kalman filter-based Metropolis-Hastings proposal densities with block-sampling schemes. Factor model uncertainty is also fully addressed by a reversible jump Markov chain Monte Carlo algorithm designed to learn about the number of common factors. Three applications, two based on synthetic Gamma and Bernoulli data and one based on real Bernoulli data, are presented in order to illustrate the flexibility and generality of the new class of models, as well as to discuss features of the proposed MCMC algorithm. This is joint work with Dani Gamerman and Esther Salazar form the Instituto de Matematica, Universidade Federal do Rio de Janeiro, Brazil.

Inici
MODELLING EXTREME TEMPERATURES AND ELECTRICITY DEMAND IN SPAIN

CONVIDAT: Dolores Furió

IDIOMA: Castellà

LLOC: Edifici C5, Aula: C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 19 de Juny de 2009, Hora: 13:00

RESUM: Much of the statistical analyses regarding weather studies use the average temperature as one of the main variables. However, as the Intergovernmental Panel on Climate Change, in its 4th assessment report states, "the type frequency and intensity of extreme events are expected to change as Earth's climate changes". In this paper the focus is on the statistical behaviour of extreme values of this random variable as well as its relationship with extreme values of electricity demand. Being able to approximate the frequencies and to identify time trends of such events is of great concern to public and private sectors, when defining strategies to adapt to climate change and, particularly, to the electricity sector since electricity is not storable. The methodology of Extreme Value Theory is employed. The generalised extreme value (GEV) distribution and the generalised Pareto distribution (GPD) are fitted to data by applying respectively the method of block maxima and the peak over threshold method.

Inici
A SIMULATION STUDY OF THE FLEET SIZING PROBLEM ARISING IN OFFSHORE ANCHOR HANDLING OPERATIONS

CONVIDAT: Irina Gribkovskaia and Aliaksandr Shyshou, Molde University College, Molde, Norway

IDIOMA: Anglès

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Dijous, 18 de juny de 2009, 12:00.

RESUM: A fleet sizing problem arising in anchor handling operations related to movement of offshore mobile units is presented in this paper. Typically, the intensity of these operations is unevenly spread throughout the year. The operations are performed by dedicated vessels, which can be hired either on the long-term basis or on the spot market. Spot rates are frequently a magnitude higher than long-term rates, and vessels are hired on the spot market if there is a shortage of long-term vessels to cover the ongoing anchor handling operations. Deciding the cost-optimal fleet of vessels on the long-term hire to cover future operations is a problem facing offshore oil and gas operators. This decision has a heavy economic impact as anchor handling vessels are among the most expensive ones. The problem is of highly stochastic nature as durations of anchor handling operations vary and depend on uncertain weather conditions. Moreover, future spot rates for anchor handling vessels are extremely volatile. The objective of this paper is to describe a simulation model for the fleet sizing problem. The study was initiated by the largest Norwegian offshore oil and gas operator and has received considerable acceptance among the planners.

Inici
SISTEMAS DINÁMICOS DE TIPO GRADIENTE

CONVIDAT: Aris Daniilidis, Departament de Matemàtiques, Universitat Autònoma de Barcelona

IDIOMA: Castellà

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 29 de maig de 2009, 12:00.

RESUM: En este seminario haremos una introducción en la teoría de los sistemas dinámicos de tipo gradiente, comentando sus propiedades características con respecto a los sistemas generales. Nos enfocaremos en particular en la teoría asintótica (convergencia de orbitas, longitud) presentando nuevos resultados que tienen orígenes en la geometría algebraica y aplicaciones en la optimización moderada.

Web del ponent: clica aquí.

Inici
ESTIMACIÓN NO PARAMÉTRICA DE INTEGRALES DE SUPERFICIE

CONVIDAT: Raúl Jiménez, Departamento de Estadística, Universidad Carlos III de Madrid.

IDIOMA: Castellà

LLOC: Edifici C5, Aula: C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 22 de Maig de 2009, Hora: 12:30

RESUM: La estimación de funcionales sobre la frontera de un cuerpo desconocido es un reto para la estadística paramétrica con útiles aplicaciones a la física, al análisis de imágenes y a la estereología. Si se es capaz de determinar cuando un disparo impacta el cuerpo, Cuevas y otros [Ann. Statist., 35:1031-1051] estiman la medida de Hausdorff de la frontera (longitud en dimensión 2, área de superficie en dimensión 3) disparando aleatoriamente a una caja que contenga al cuerpo. Los estadísticos considerados por estos autores y en otros trabajos subsiguientes están basados en la estimación del contenido de Minkowski y dependen de un parámetro de suavización que debe ser escogido cuidadosamente. Para el mismo esquema muestral, discutiremos un enfoque no paramétrico sin esta incomodidad que permite estimar tanto la medida de Hausdorff como integrales de superficie de una función escalar, cuando podemos muestrear los valores de la función en la puntos de impacto. Ejemplos de este tipo de datos provienen de experimentos en los cuales la densidad de un cuerpo puede ser medida por propiedades físicas de los impactos o de escenarios en los cuales cantidades, tales como temperatura y humedad, son observados por sensores aleatoriamente distribuídos. El método que discutiremos es el primero en considerar el problema de estimación de integrales de superficie bajo este tipo de modelos muestrales. Está basado en triangulaciones aleatorias de Delaunay e incluye un procedimiento simple de reconstrucción de superficies a partir de nubes densas de puntos fuera y dentro del cuerpo. Se discutirán resultados asintóticos, simulaciones y aplicaciones.

 

Cliqueu aquí per accedir a la pàgina web d'en Raúl Jiménez

Inici
PLANIFICACIÓN EFICIENTE DE SISTEMAS TERRITORIALES MEDIANTE UN MÉTODO AVANZADO DE LOCALIZACIÓN, ASIGNACIÓN Y BÚSQUEDA LOCAL

CONVIDAT: Roger Z. Ríos, Programa de Posgrado en Ing. de Sistemas, Universidad Autónoma de Nuevo León, México

IDIOMA: Castellà

LLOC: Aula C5016, Campus Nord, UPC (veure mapa)

DATA: Divendres, 24 d'abril de 2009, 12:00.

RESUM: Los problemas de diseño territorial consisten en agrupar pequeñas unidades geográficas en grupos de mayor tamaño, denominados territorios o distritos, de tal forma que éstos satisfacen ciertos criterios relevantes de planificación.  Estos requerimientos pueden ser motivados económicamente (potencial de ventas, carga de trabajo, número de clientes) o demográficamente (número de habitantes, población votante.  Además, se imponen con frecuencia restricciones espaciales, como compacidad o conectividad territorial.  Los problemas de diseño territorial aparecen en diversas aplicaciones como diseño de distritos electorales, diseño de distritos escolares, diseño de territorios de ventas, diseño de planes de reciclaje de aparatos electrodomésticos, por mencionar algunas. En particular, en esta charla se presenta un problema de diseño territorial motivado por una aplicación práctica en la distribución de bebidas embotelladas. El problema consiste en encontrar un conjunto dado de territorios que satisfagan criterios de compacidad, conectividad y requerimientos múltiples de balance con respecto al número de clientes y demanda del producto.  Se discutirá la descripción y modelado del problema como un Programa Entero Mixto Lineal y se describirá en detalle una novel metodología de solución basada localización-asignación-búsqueda local para este problema NP-duro.  La charla incluye una exhaustiva evaluación numérica del método propuesto en una variedad de instancias que demuestra su excelente comportamiento y una discusión sobre posibles extensiones a una clase más amplia de sistemas territoriales no explorados previamente.

TRANSPARENCIES DE LA PRESENTACIO: clica aquí

Inici
UNMIXING COMPOSITIONAL DATA WITH MCMC

CONVIDAT: Raimon Tolosana. Raimon Tolosana-Delgado (Ph.D.) Dept. d'Enginyeria Hidràulica, Marítima i Ambiental Laboratori d'Enginyeria Marítima (LIM/UPC) Universitat Politècnica de Catalunya.

IDIOMA: Català

LLOC: Seminari EIO, ETSEIB (Edifici Eng. Industrial), Planta 6, Campus Sud, Universitat Politècnica de Catalunya Avda. Diagonal, 647

DATA: Divendres, 17 d'abril de 2009, 12:30.

RESUM: An end-member problem takes some multivariate observations as generated by a composition (a convex linear mixture) of a few "pure" members, and estimates these end-member characteristics as well as the coefficients of the linear mixture for our observations. In this talk, the problem is motivated by trying to re-express the geochemical composition (observable) of a series of sediment samples into a composition of minerals (the end-members). First, we assume some fixed characteristics of the end-members, and study the set of solutions from a purely geometric perspective. Second, we discuss how to select "best" solutions from this space, in particular under some distributional assumptions. Third, we allow the end-member properties to vary in a controlled, interpretable fashion. Finally we build a Bayesian model, with some parameters characterizing each of these three steps, that can be estimated by conventional MCMC techniques.

 Inici