Programació del seminari: Any 2015
- Dijous 3 de desembre de 2015, Hora 12:00
Using integer programming and decomposition to tackle real-time train traffic management
Leonardo Lamorgese, SINTEF Applied Mathematics, Oslo,Noruega.
- Divendres 6 de novembre de 2015, Hora 12:30
Bayesian estimations of bandwidth in semiparametric discrete kernel estimations
Célestin Kokonendji, Laboratoire de mathématiques de Besançon, Université de Franche-Comté.
- Divendres 30 d'octubre de 2015, Hora 12:30
Modelos de regresión multivariante en estudios observacionales publicados en revistas biomédicas: ¿Se está reportando bien su validación?
Jose M Martínez-Sánchez, Unidad de Investigación y Control del Tabaquismo del Institut Català d'Oncologia y Universitat Internacional de Catalunya, Barcelona.
- Divendres 26 de juny de 2015, Hora 12:30
Bayesian Approaches for Early Phase Clinical Trials
Pantelis Vlachos, Cytel inc., Geneva branch, Switzerland.
- Divendres 19 de juny de 2015, Hora 12:30
Classification methods for Hilbert data based on surrogate density
Aldo Goia, Dipartimento di Studi per l'Economia e l'Impresa, Università del Piemonte Orientale "A. Avogadro", Novara, Italy.
- Divendres 12 de juny de 2015, Hora 12:30
A Closed-loop Approach to Dynamic Assortment Planning
Víctor Martínez de Albéniz, Departament de direcció de producció, tecnologia i operacions, IESE, Bareclona.
- Divendres 27 de febrer de 2015, Hora 12:30
Herramientas bioinformáticas en ecología microbiana
Marcelo Soria, Cátedra de Microbiología, Facultad de Agronomía, Universidad de Buenos Aires, Buenos Aires, Argentina.
- Divendres 9 de gener de 2015, Hora 12:30
An empirical comparison of methods to meta-analyze individual patient data of diagnostic accuracy
Gabrielle Simoneau, BSc, Department of Epidemiology, Biostatistics and Occupational Health, McGill University, Montreal Chest Institute, McGill University Health Centre, Montreal, Canada.
Inici
Using integer programming and decomposition to tackle real-time train traffic management
CONVIDAT: Leonardo Lamorgese
IDIOMA: Anglès
LLOC: Edifici C5, Aula C5016, Campus Nord, UPC (veure mapa)
DATA: Dijous, 3 de desembre de 2015. Hora: 12:00
RESUM: Trains moving in railway systems are often affected by delays, disturbances or cancellations. These undesired events may be alleviated by suitably re-routing and re-scheduling trains, in real-time. Such task (also called train dispatching) is central in managing reilway systems as it allows recovering from undesirable deviations from the timetable and better exploiting railway resources. With few exceptions, dispatching is still almost entirely in the hands of human operators, despite the problem amounts to solving a complex and large optimization problem. In this talk, we describe how integer programming can be exploited to find optimal solutions within the stringent time limit requied by this application. In particular, we show how to decompose the problem into smaller sub-problems, associated with different parts of the network. This decomposition is the basis for a master-slave solution algorithm, in which the two sub-problems are modeled as mixed integer linear programs, with specific sets of variables and constraints. Similarly to the classical Benders' decomposition approach, slave and master communicate through suitable feasibility cuts in the variables of the master.
A decision support system based on this exact approach was put in operation in Norway in February 2014 and represents, to our knowledge, the first operative application of mathematical optimization to train dispatching. Other dispatching systems based on heuristic versions of our algorithms for main line and large stations are in operation in Italy and Latvia.
EL PONENT: Leonardo Lamorgese és investigador al departament de Matemàtica Aplicada de SINTEF a Oslo, Noruega. És graduat per la universitat de Roma La Sapienza en Decision Science and Operations Research. Els seus temes d'interès i la seva recerca s'enfoquen a aplicacions de la Matemàtica Discreta al transport ferroviari, planificació d'esports i gestió sanitària. La seva recerca s'ha publicat en revistes com Operations Research i Transportation Science. El 2014, AIRO (societat italiana d'investigació operativa) li va atorgar el reconeixement internacional "Best Application 2014" pel seu treball "An exact decomposition approach for the Train Dispatching Problem".
Clica aquí per accedir a la seva pàgina web a SINTEF
Inici
Bayesian estimations of bandwidth in semiparametric discrete kernel estimations
CONVIDAT: Célestin Kokonendji
IDIOMA: Anglès
LLOC: Edifici C5, Aula C5016, Campus Nord, UPC (veure mapa)
DATA: Divendres, 6 de novembre de 2015. Hora: 12:30
RESUM: Reminding the framework of discrete smoothing using discrete associated kernel methods by distinguishing categorial to count data, I will present three Bayesian approaches to selecting bandwidth: global, local and adaptive. Each one of these techniques will be discussed and compared for binomial kernel. Then, I will give tools of semiparametric discrete kernel estimations of a probability mass function under the Poisson-weighted assumption; model diagnostics will be also evoked. Some applications on real data will conclude this talk.
EL PONENT: Célestin Kokonendji is full professor in Statistics and Probability at the Laboratoire de Mathématique de Besançon of the Université de Franche-Comté. His research interests concern, among others, Associate Kernel Methods (Discrete/Continuous/Mixed), Nonparametric Statistics, Bayesian Analysis, Models and Analysis of Over/Underdispersed Count Data, Environmental Statistics, Statistical Inference, Distribution Theory, Dispersion Models, Generalized Variance Functions, Monge-Ampère equation, Pseudo-Orthogonal Polynomials, Stochastic/Lévy Processes and Applied Stochastic Models. He is associate editor of various statistical journals, has directed more than 10 PhD students and (co)-authored over 60 scientific articles. For more information on his research, click on http://lmb.univ-fcomte.fr/celestin-c-kokonendji
Inici
Modelos de regresión multivariante en estudios observacionales publicados en revistas biomédicas: ¿Se está reportando bien su validación?
CONVIDAT: Jose M Martínez-Sánchez
IDIOMA: Castellà
LLOC: Edifici C5, Aula C5016, Campus Nord, UPC (veure mapa)
DATA: Divendres, 30 de octubre de 2015. Hora: 12:30
RESUM: Dos aspectos importantes en la investigación biomédica son la validez interna y externa de los estudios. El sesgo de información y la confusión afecta a la validez interna de los estudios y están presentes hasta cierto punto en toda investigación observacional. El sesgo de confusión puede ser corregido, siempre y cuando el factor de confusión sea anticipado y se haya recogido la información. La confusión puede ser controlada de diversas formas, siendo los modelos de regresión multivariante (MRM), tales como regresión lineal, logística, Poisson, o de Cox, los más utilizados y popularizado en la investigación biomédica en los últimos años. Sin embargo, los MRM asumen ciertos supuestos (normalidad, homocedasticidad, independencia de los errores, etc.) que de no ser cumplidos las estimaciones de los parámetros pueden estar segadas aumentado el posible error tipo I de las estimaciones. El este trabajo se revisa la calidad del reporte de la aplicación de los MRM más utilizados (regresión logística, lineal y Cox) en estudios observacionales analíticos publicados en revistas indexadas en PubMed entre 2003 y 2014.
EL PONENT: El Dr. Jose M Martínez-Sánchez, es Doctor por la Universidad de Barcelona (Premio Extraordinario de Doctorado y Mención Doctorado Europeo), Master en Salud Pública por la Universidad Pompeu Fabra y Estadístico por la Universidad de Extremadura. La actividad investigadora y profesional la realiza en la Unidad de Investigación y Control del Tabaquismo del Institut Català d'Oncologia (Centro colaborador de la OMS en control del tabaquismo) en calidad de epidemiólogo e investigador. Además, es el responsable del área de bioestadística de la Universitat Internacional de Catalunya. El Dr. Martínez-Sánchez ha sido investigador (principal y colaborador) en numerosos proyectos nacionales e internacionales. Sus principales líneas de investigación son la evaluación de las medidas del control del tabaquismo y los estudios sobre consumo de tabaco y exposición al humo ambiental del tabaco en población general mediante el uso de cuestionario y biomarcadores (cotinina). Su producción científica incluye más de 60 artículos indexados en Science Citation Index (Thomson-ISI).
Inici
Bayesian Approaches for Early Phase Clinical Trials
CONVIDAT: Pantelis Vlachos
IDIOMA: Anglès
LLOC: FME, Edifici U, Aula: Sala de Juntes, Campus Sud, UPC, Pau Gargallo, 5, 02028 Barcelona (veure mapa).
DATA: Divendres, 26 de juny de 2015. Hora: 12:30
RESUM: In this work, we evaluate through simulations the operating characteristics of Bayesian Adaptive Randomization (BAR) in proof-of-concept dose selection studies as well as the Continual Reassessment method for Phase I dose escalation studies in oncology.
In the former, the statistical design links treatment assignment probabilities to performance of respective arms. A Bayesian model summarizing prior information has been implemented with priors assuming no activity in order not to unbalance the randomization. The posterior probability that placebo has better performance than experimental treatment arms will be used to reject a null hypothesis of no drug activity. A screening design was used to calculate the maximum sample size and to compare operating characteristics of the two methods.
In the latter we utilize pre-clinical knowledge to obtain a characterization of the dose-toxicity relationship and combine this information with observed data to estimate the maximum tolerated dose in an optimum way (by minimizing the chance of under/over dosing). We compare this approach to standard rule-based approaches in Oncology.
EL PONENT: Pantelis Vlachos is Director of Strategic Consulting at Cytel. He is responsible for assisting pharmaceutical and biotech clients with the design of their clinical development programs and clinical trials, including adaptive designs of late stage trials and early stage dose-finding trials. He participates in designing, development, testing and refinement of Cytel's software products. Prior to Cytel, he was principal biostatistician at Merck-Serono where, among other activities, he developed and implemented standard analytical and graphical tools for reporting of safety data. Pantelis was also a Professor of Statistics at Carnegie Mellon University for 12 years. His research interests lie in the area of adaptive designs, mainly from a Bayesian perspective, as well as hierarchical model testing and checking although his secret passion is Text Mining. He is the Managing Editor of the journal “Bayesian Analysis” and has served in the editorial boards of several other journals and online statistical data and software archives.
Inici
Classification methods for Hilbert data based on surrogate density
CONVIDAT: Aldo Goia
IDIOMA: Anglès
LLOC: C4002
DATA: Divendres, 19 de juny de 2015. Hora: 12:30
RESUM: We study classification approaches for Hilbert random curves resting on the use of a surrogate of the probability density which is defined, in a distribution-free mixture context, from an asymptotic factorization of the small-ball probability. The latter is rigorously established exploiting the Karhunen-Loéve expansion whose basis turns out to be the optimal one in controlling the approximation errors. That surrogate density is estimated by a kernel approach from the principal components of the data. The remaining part of the work focuses on the illustration of the classification algorithms and the computational implications, with particular attention to the tuning of parameters involved. Some asymptotic results are sketched. Applications on simulated and real datasets show how the proposed methods work.
EL PONENT: Aldo Goia is Associate Professor of Statistics in the Dipartimento di Studi per l'Economia e l'Impresa, Università del Piemonte Orientale "A. Avogadro", Italia. He obtained his PhD in 2003 at the Université Paul Sabatier (Toulouse - France) under the supervision of Philippe Vieu and P. Sarda. His research topics are Functional Data Analysis (FDA), Multivariate Statistics and Goodness-of-fit tests. He has published papers on the single functional index model, functional projection pursuit, functional clustering and functional time series. He is interested in methodology, but also in applications of FDA.
Inici
A Closed-loop Approach to Dynamic Assortment Planning
CONVIDAT: Victor Martínez de Albéniz
IDIOMA: Castellà
LLOC: Edifici C5, Aula C5016, Campus Nord, UPC (veure mapa)
DATA: Divendres, 12 de juny de 2015. Hora: 12:30
RESUM:
Firms are constantly trying to keep the customers interested by refreshing their assortments. In industries such as fashion retailing, products are becoming short-lived and, without product introductions or in-store novelties, category sales quickly decrease. We model these dynamics by assuming that products lose their attractiveness over time and we let the firm enhance the assortment at a cost, for single or multiple categories. We characterize the optimal closed-loop policy that maximizes firm profits. When adjustment costs are linear in the attractiveness, we find that an assort-up-to policy is best: it is optimal to increase
category attractiveness to a target level, which is independent of the current attractiveness. Interestingly, when there is no constraint on the maximum achievable attractiveness for any category, it is optimal to invest in one category only, in the extreme cases single period and infinite horizon. In that case the optimal assort-up-to levels can be characterized in closed-form. In contrast, with capacity constraints or multi-period finite horizons, it is optimal to diversify the investment across categories. Beyond our structural results, we study the value of our closed-loop approach compared to open-loop and front-loaded strategies. We find that our policies can provide significant value (up to 10% of additional profits), especially when the costs are not too high nor too low, and there is significant uncertainty about the decay rate of products, as in most fashion retailing contexts.
This is a joint work with Esra Çinar.
EL PONENT:
Víctor Martínez de Albéniz is associate professor in IESE’s Department of Production, Technology and Operations Management.
He joined IESE in 2004 after earning a Ph.D. at the Operations Research Center of the Massachusetts Institute of Technology (MIT) and an engineering degree at École Polytechnique in France.
His research focuses on supply chain management, where procurement, production and distribution decisions can help companies compete more successfully in the global arena. In particular, he has studied in depth procurement and supply issues, where a balanced sourcing portfolio can provide low cost, flexibility and innovation opportunities. More recently, he has worked on retail topics, especially in the fashion industry. He has published his work in journals such as Management Science, Operations Research, Manufacturing and Services Operations Management, or Production and Operations Management.
In addition, Prof. Martínez de Albéniz teaches IESE courses on operations management, operations strategy, advanced methods for operations and new product development, both at the executive and MBA levels. He has also taught at other schools such as MIT, MDE (Côte d’Ivoire) or the Indian School of Business.
Per més informació, visita la seva pàgina web http://blog.iese.edu/martinezdealbeniz/
Inici
Herramientas bioinformáticas en ecología microbiana
CONVIDAT: Marcelo Soria
IDIOMA: Castellà
LLOC: Edifici C5, Aula C5016, Campus Nord, UPC (veure mapa)
DATA: Divendres, 27 de febrer de 2015. Hora: 12:30
RESUM: ¿Cuántas bacterias diferentes hay en un gramo de suelo? ¿Cuántos hongos? ¿Importa saberlo? ¿Cuántos microorganismos viven dentro de nuestros intestinos? ¿Son peligrosos? Estas no son preguntas nuevas para los microbiólogos especializados en ecología. En este momento la diferencia es que gracias a avances tecnológicos, sobre todo en metodologías para secuenciar ADN, sabemos que las comunidades de microorganismos son mucho más grandes y complejas de lo que suponíamos. También estamos descubriendo que estas comunidades son dinámicas y sus modificaciones se asocian a cambios, por ejemplo, en la fertilidad de los suelos o en nuestra salud. Los estudios ecológicos modernos se distinguen por la gran cantidad de datos que generan. La mayor parte de éstos son moleculares, y los equipos de investigación necesitan incluir un bioinformático para procesarlos. El análisis de datos en esta área sigue los principios generales que se aplican en cualquier trabajo de minería de datos, y también se incorporan métodos y conceptos propios del dominio biológico. En este seminario recorreremos el trabajo típico de un bioinformático dedicado a la ecología microbiana. También revisaremos el impacto de estos estudios en agricultura y medicina. Y plantearemos cuáles son los próximos desafíos que tendrán que enfrentar los bioinformáticos en esta área.
EL PONENT: Marcelo Soria es profesor en la Cátedra de Microbiología de la Facultad de Agronomía de la Universidad de Buenos Aires. Es Doctor en Ciencias Biológicas por la Universidad de Buenos Aires. Es (co-)autor de unas 35 publicaciones en revistas científicas, majoritariamente internacionales. Ha dirigido varias tesis de llicenciatura, de maestría y de doctorado.
An empirical comparison of methods to meta-analyze individual patient data of diagnostic accuracy
CONVIDAT: Gabrielle Simoneau
IDIOMA: Anglès
LLOC: Edifici C5, Aula C5016, Campus Nord, UPC (veure mapa)
DATA: Divendres, 9 de gener de 2015. Hora: 12:30
RESUM: Individual patients data (IPD) meta-analysis are more and more common in the literature. The use of IPD has many benefits. In the context of diagnostic accuracy studies, pooled sensitivity and specificity are traditionally reported for a given threshold and meta-analysis are conducted via a bivariate approach. More precisely, the bivariate random-effects model (BREM) accounts for the correlation between the two endpoints and has shown to produce satisfying results. With IPD, it is possible to obtain pooled sensitivity and specificity for each possible threshold. One way to analyze these data is to apply the BREM model at every threshold. Another idea is to analyze results for all thresholds simultaneously. The correlation between results for different cutpoints within a study is then taken into account. Several approaches have been proposed to do this including the multivariate random effects model (MREM) and the mixed ordinal model. Our aim is to compare these two approaches to the BREM when IPD are available, empirically. Ordinal and semi-continuous scales are considered in this analysis.
LA PONENT: Gabrielle Simoneau is a 2nd year MSc student in Biostatistics at McGill University, Montreal (CAN). She obtained her bachelor in Mathematics and Statistics at Université de Montréal in June 2013. She has worked on large data classification methods under the supervision of Dr.Xavier Bry at Université Montpellier II (France). She is currently writing her master thesis under the supervision of Dr.Andrea Benedetti, which explores the benefits of individual patients data (IPD) in the context of test accuracy diagnostic studies. Besides her research, she occasionally collaborates as a statistical analyst in various medical studies and works as a statistical consultant.
Comparteix: