Oradores Invitados

Natalia da Silva (PhD, Iowa State University): Es profesora adjunta en el Instituto de Estadística de la Universidad de la República en Montevideo (UDELAR-IESTA). Trabaja en investigación en métodos de aprendizaje supervisado, predicción, análisis de datos exploratorios y gráficos estadísticos en colaboración con Di Cook y Heike Hofmann. Es cofundadora de R-Ladies Ames y R-Ladies Montevideo y Chair de LatinR2018.
Visualización estática e interactiva de datos usando ggplot2 y plotly
La visualization tiene un rol fundamental en todas las etapas del análisis estadístico de datos, exploración, modelado y diagnóstico. Permite descubrir patrones escondidos en los datos así como dar luz a modelos o algoritmos complejos. En este taller vamos a recorrer los conceptos básicos para hacer visualización estática de datos con R usando el paquete ggplot2. Veremos las ideas fundamentales de la gramática gráfica que está detrás de ggplot2. Los participantes aprenderán a crear distintos tipos de gráficos en pocos pasos de calidad publicable. A su vez Introduciremos el concepto de gráficos interactivos y su potencial para el descubrimiento de patrones y el análisis estadístico de datos. Existen diversas opciones para hacer visualización interactiva en R pero en este taller usaremos plotly. 
Todos los tutoriales requieren traer una notebook con la batería cargada y que tenga instalados la última versión de R y RStudio al 1 de Junio de 2018. También necesitará tener instalado el paquete tidyverse y plotly.
 
Andrés Farrall: Especialista en temas relacionados a la Ciencia de Datos. Actualmente es el Responsable del Área Metodológica de Ecoclimasol y Profesor de Postgrado en la Facultad de Cs. Exactas y Naturales (UBA).
Inteligencia Artificial con R: Introducción al “Deep Learning”
El taller mostrará herramientas de “Deep Learning” aplicadas a problemas de Regresión, Clasificación y “Novelty Detection”. Se utilizará la interfaz KERAS basada en TensorFlow, tecnología de código abierto creada por Google para el desarrollo del aprendizaje automático. Se enseñarán a lo largo del taller los conceptos fundamentales de los métodos tratados, y se darán ejemplos de caso mediante el uso del entorno de desarrollo integrado Rstudio. 
Todos los tutoriales requieren traer una notebook con la batería cargada y que tenga instalados la última versión de R y RStudio al 1 de Junio de 2018. También necesitará tener instalado el paquete tidyverse.
 
Jennifer Bryan (PhD, UC Berkeley): Ingeniera de Software en RStudio y Profesora Asociada de la University of British Columbia. Es una referente internacionalmente reconocida de la comunidad R dentro de la cual es integrante ordinaria de la R Foundation y parte del Comité de Liderazgo de rOpenSci.
How to repeat yourself with purrr
Iterative tasks come up often in data analysis: do X for every Z, where Z could be 'column', 'row', 'treatment group', etc. In base R, the nicest way to attack these tasks is with the 'apply' functions. The 'apply' functions are very powerful, but also have some downsides: they do not present a consistent interface and they often return data in an awkward form. The tidyverse is an opinionated collection of R pckages designed for data science and the purrr package, specifically, is extremely useful for doing repetitive tasks. purrr's 'map' functions have an extremely consistent interface and return data in a form that is ready for the next step in your analysis. 
This tutorial includes:
Brief orientation to the tidyverse.
Brief review of how write a basic function. If you've never written a function before, see the three parts of this to prepare. In-depth coverage, both conceptual and hands-on, of the purrr package.
Case studies, such as how to do work on each row in a data frame or how to turn data in a nested list (e.g. from JSON or XML) into a nice rectangle.
 
Teaching R (and more) in the Era of Data Science
My talk deals with two related themes: the ongoing discussion of 'data science vs. statistics' and the importance of developing your own data analysis workflow. These topics are related in my mind because I believe, as academic statisticians, we often take an unnecessarily narrow view of our discipline. The 'data science vs statistics' debate brings this to a head, because the desire to join and even lead data science initiatives provides an incentive to broaden our mandate. What if we embraced the development and teaching of tools -- both mental and digital -- that address the entire data analysis process? I'll draw on my experience developing and delivering STAT 545 at the University of British Columbia. I conclude with an overview of semi-recent developments in the R ecosystem, aimed at people who want to make their workflow more productive and less aggravating. I'll explain what the tidyverse is and why so many people are incorporating new tools, like Git and GitHub, into their workflow.
Todos los tutoriales requieren traer una notebook con la batería cargada y que tenga instalados la última versión de R y RStudio al 1 de Junio de 2018. También necesitará tener instalado el paquete tidyverse.
 

Walter Sosa Escudero: Investigador Principal de CONICET. Profersor y Director del departamento de Economía - Universidad de San Andrés. TEDxRiodelaPlata. Autor de ¿Qué es (y qué no es) la estadística? y El lado oscuro de la econometría.

Aprendiendo a computar vs computar para aprender: hacia una nueva forma de enseñar y aprender estadística
La revolución de datos y algoritmos ofrece una excelente oportunidad para renovar la forma en la que se motiva y enseña la estadística. Lejos de funcionar como una mera ilustración de los métodos formales, la charla sugiere que entornos computacionales flexibles como R pueden cumplir un rol fundamental en explicar el funcionamiento de ideas complejas en el campo del análisis moderno de datos.

 

Octavio Lange Ingeniero Agrónomo (UBA) y Magíster en Administración de Negocios (UNICEN): Nacido en Tandil, con más de 6 años especializándose en tecnologías digitales en el sector agropecuario. Asesor CREA de la región Mar y Sierras y responsable del proyecto Palenque de la Fundación Sadosky. 

El proyecto Palenque como facilitador del ecosistema AgTech.

 Día tras día surgen nuevas y mejores herramientas para agregar valor o resolver problemáticas del. sector agropecuario, tanto públicas como privadas. El. inconveniente es que cada actor brinda soluciones estancas con bajo grado de colaboración entre ellas. El. desafío es acordar un lenguaje común y lograr que estas herramientas puedan interoperar entre sí para alcanzar mejores soluciones y generar mayor conocimiento colectivo y conectivo

 

Prof. Dr.-Ing. Armando Walter Colombo (Fellow IEEE): joined the Department of Electrotechnical and Informatics at the University of Applied Sciences Emden-Leer, Germany, became Full Professor for Industrial Informatics in August 2010 and Director of the Institute for Industrial Informatics, Automation and Robotics (I2AR) in 2012. He worked during the last 17 years as Manager for Collaborative Projects and also as Edison Level 2 Group Senior Expert at Schneider Electric, Industrial Business Unit.  He received the BSc. on Electronics Engineering from the National Technological University of Mendoza, Argentina, in 1990, the MSc. on Control System Engineering from the National University of San Juan, Argentina, in 1994, and the Doctor degree in Engineering (Production Automation and Systematization) from the University of Erlangen- Nuremberg, Germany, in 1998. From 1999 to 2000 was Adjunct Professor in the Group of Robotic Systems and CIM, Faculty of Technical Sciences, New University of Lisbon, Portugal. His research interests are in the fields of industrial cyber-physical systems, industrial digitalization and system-of-systems engineering, Internet-of-Services, Industry 4.0- compliant solutions. Prof. Colombo has over 30 industrial patents and more than 300 per-review publications in journals, books, chapters of books and conference proceedings (see https://scholar.google.com/citations?user=csLRR18AAAAJ). Prof. Colombo has extensive experience in managing multi-cultural research teams in multi-regional projects. He has participated in leading positions in many international research and innovation projects in the last 16 years. With his contributions, he has performed scientific and technical seminal contributions that are nowadays being used as one of the basis of what is recognized as “The 4th Industrial Revolution”: networked collaborative smart cyber-physical systems that are penetrating the daily life, producing visible societal changes and impacting all levels of the society. He is co-founder of the IEEE IES TC on Industrial Agents and TC on Industrial Informatics. He is founder and currently Chairman of the IEEE IES TC on Industrial Cyber- Physical Systems and member of the IEEE IES Administrative Committee (AdCom). Prof. Colombo served/s as advisor/expert for the definition of the R&D&I priorities within the Framework Programs FP6, FP7 and FP8 (HORIZON 2020) of the European Union, and he is working as expert/evaluator in the European Research Executive Agency (REA), ECSEL, Eureka- and German BMBF/DLR IKT-Programs. Prof. Colombo is listed in Who’s Who in the World /Engineering 99-00/01 and in Outstanding People of the XX Century (Bibliographic Centre Cambridge, UK).

Working with Industrial Cyber-Physical System and Industry4.0-compliant Solutions (Education and Training Requirements)

We are witnessing rapid changes in the industrial environment, mainly driven by business and societal needs towards production customization and the digitalization of the economy, i.e., digitalization and interconnection of products, services, enterprises, and people. This trend is supported by new disruptive advances in the cross-fertilization of concepts and the amalgamation of information-, communication-, control- and mechatronics technology- driven approaches in traditional industrial systems. In this context, industrial informatics combine the progress achieved by the application of large distributed and networked computing systems on product and production system design, planning, engineering, and operation with the power of digital data that are produced by industrial processes and also collected by the Internet of Things. The technological, economic, and social impacts of these developments are so enormous that the whole process is labeled as the 4th Industrial Revolution.

In 2006, the term “Cyber-Physical Systems” (CPS) was coined to “refer to the integration of computation with physical processes”. CPS can be described as smart systems that encompass hardware and software, computational and physical components, seamlessly integrated and closely interacting to sense and to control in real-time the changing state of the real world. These systems involve a high degree of complexity at numerous spatial and temporal scales, and highly networked communications integrating their computational and physical components. As such CPS refer to Information-Communication-Control- Mechatronics-Systems (sensing, actuating, computing, communicating, etc.) embedded in physical objects, interconnected through several networks including the Internet, and providing citizens and businesses with a wide range of innovative applications based on digitalized data, information and services.

Ontologically the term Cyber-Physical Systems means hardware-software systems which tightly couple the physical world and the digitalized (virtual) world. In a CPS ecosystem, on the one hand every real physical object has one or more cyber representations, and on the other hand a cyber component or system can be linked to a physical representation i.e., an object in the 3-dimensional human-tangible world. Moreover, these objects are increasingly interconnected in real-operational-time, networked either permanently or communicate in an asynchronous manner from time to time, possessing all essential characteristics of real-time-critical systems.

Industrial Cyber-Physical Systems (ICPS) and Industry4.0-compliant solutions forge the core of real-world networked industrial infrastructures having a cyber-representation through digitalization of data and information across the enterprise, along the product and process engineering life-cycle and from suppliers to customers along the supply chain. As such the competitive performance of an ICPS mainly depends on the ability to effectively collect, analyze and use large-scale digitalized data and information from many different and often heterogeneous sources, to sustainably and efficiently manage, supervise and operate in the industrial environments. This effective information-driven interaction of ICPS with other CPS and enterprise systems, extending to all business processes, is viewed as vital to modern Industry4.0-compliant infrastructures.

There are many challenges ahead in the convergence of computing, control, mechatronics and communications for CPS and Industry4.0-compliant ecosystems. There is a need for investigating and learning a wide spectrum of foundations, research and technological fields. In this context, the plenary talk (i) addresses the penetration and proliferation of such ecosystems into the industrial environments, taking into account that the same trend is also evident in other domains such as energy, healthcare, manufacturing, military, transportation, consumer, enterprise, robotics, and smart cities, among others; and (ii) offers an overview of major requirements to learning, teaching, training students, technicians and engineers for working with ICPS and Industry4.0-compliant solutions.

After presenting the scientific and technical background behind Industrial Cyber-Physical Systems (ICPS), their relationship to other frameworks like Systems-of-Cyber-Physical Systems, Industrial Internet-of-Things (IIoT) and Industry4.0, and associated technology standardization/normalization initiatives, the audience/participants of the plenary talk will get a deep view about:

● Which is the minimal necessary pre-existing Know-How for understanding and working with ICPS and Industry4.0-compliant solutions?

● What and How to learn ICPS and Industry4.0-compliant solutions? (Recommendations for graduated and post-graduate students, and industrialists)

● What and How to teach ICPS and Industry4.0-compliant solutions? (Recommendations for trainees, professors, etc.)

● How to engineer and operate ICPS and Industry4.0-compliant solutions?

 

Prof. Ignacio E. Grossmann: is the R. R. Dean University Professor of Chemical Engineering, and former Department Head at Carnegie Mellon University. He obtained his B.S. degree at the Universidad Iberoamericana, Mexico City, in 1974, and his M.S. and Ph.D. at Imperial College in 1975 and 1977, respectively.  He is a member and former director (2005-2015) of the “Center for Advanced Process Decision-making,” an industrial consortium that involves about 20 petroleum, chemical, engineering and software companies. He is a member of the National Academy of Engineering. He has received the following AIChE awards, Computing in Chemical Engineering, William H. Walker for Excellence in Publications, Warren Lewis for Excellence in Education, and Research Excellence in Sustainable Engineering. In 2003 he received the INFORMS Computing Society Prize, and is currently a Fellow of INFORMS. In 2015 he was the first recipient of the Sargent Medal by the IChemE. He has honorary doctorates from Abo Akademi in Finland, University of Maribor in Slovenia, Technical University of Dortmund in Germany, University of Cantabria in Spain, and from the Russian Kazan National Research Technological University. He has been named Thomson Reuters Highly Cited Researcher in 2014-2016. His research interests are in the areas of mixed-integer, disjunctive and stochastic programming, energy systems including petroleum, shale gas and biofuels, water networks, and planning and scheduling for enterprise-wide optimization, and reliability optimization. He has authored more than 500 papers, several monographs on design cases studies, and the textbook “Systematic Methods of Chemical Process Design,” which he co-authored with Larry Biegler and Art Westerberg.  He has graduated 58 Ph.D. and 16 M.S. students.

Recent Theoretical and Computational Advances in the Optimization of Process Systems under Uncertainty

Optimization under uncertainty has been an active and challenging area of research for many years. However, its application in Process Systems has faced a number of important barriers that have prevented its effective application. Barriers include availability of information on the uncertainty of the data (ad-hoc or historical), determination of the nature of the uncertainties (exogenous vs. endogenous), selection of an appropriate strategy for hedging against uncertainty (robust optimization vs. stochastic programming), handling of nonlinearities (most work addresses linear problems), large computational expense (orders of magnitude larger than deterministic models), and difficulty in the interpretation of the results by non-expert users.

In this lecture, we describe recent advances that address some of these barriers. We first describe the basic concepts of robust optimization, including the robust counterpart, showing its connections with semi-infinite programming. We also we explore the relationship between flexibility analysis and robust optimization for linear systems. A historical perspective is given, which shows that some of the fundamental concepts in robust optimization have already been developed in the area of flexibility analysis in the 1980s. We next consider two-stage and multi-stage stochastic programming in the case of exogenous parameter, for which we describe acceleration techniques for Benders decomposition, hybrid sub-gradient/cutting plane methods for Lagrangean decomposition, and sampling techniques. We address both mixed-integer linear and nonlinear stochastic programs, including integer recourse. We then address the generalization to the case of both exogenous and endogenous parameters, which gives rise to conditional scenario trees for which theoretical properties are described to reduce the problem size. To avoid ad-hoc approaches for setting up the data for these problems, we describe approaches for handling of historical data for generating scenario trees. Finally, we illustrate the application of each of these formulations in demand-side management optimization, planning of process networks, chemical supply chains under disruptions, planning of oil and gas fields, and optimization of process networks, all of them under some type of uncertainty.

 

Dr. Cristián E. Cortés: is Associate Professor at the Civil Engineering Department, Universidad de Chile. He got the Civil Engineering title and the MSc degree from Universidad de Chile in 1995, and the PhD degree in Civil Engineering from University of California at Irvine in 2003. His areas of interest are network flows, optimization, logistics, public transport operations, equilibrium models, simulation, freight transportation, scheduling, stochastic and dynamic problems with applications in transportation. He has published 41 papers in ISI indexed journals. In addition, he has published more than 80 articles in Conference Proceedings plus other 10 papers in other non ISI journals. He has published one book plus two book chapters. He has supervised more than 40 theses at undergraduate, MSc and PhD levels, most of them at Universidad de Chile. He has worked as leader in several applied projects in the areas of logistics and transportation, including air transportation, specifically the ground handling operations of an international airport, routing and network design strategies in the distribution of e-commerce products, production and dispatch of ready-mix concrete mixers, timetabling and vehicle scheduling in public transport operations, demand responsive schemes for solving mobility issues in the case of a Chilean city, development of simulation schemes for public transport as well as distribution of services,  etc. Nowadays, he is the Chief of the Transport Engineering Division in the Civil Engineering Department at Universidad de Chile. He is also Associate Editor of Transportation Science. He has been involved in the organization of many international workshops in transport and logistics topics, being the most important his role as Chair of the Organizing Committee of the Triennial Symposium of Transportation Analysis (TRISTAN VIII) organized in San Pedro de Atacama, Chile in 2013.

GRASP heuristic scheme for solving a real case of a ready-mixed concrete production and dispatching problem

Concrete is one of the most important raw materials in the construction industry. It is commonly used in infrastructure works and buildings, where it is used from foundations to roofs. Concrete is manufactured in dedicated sites we will call hereinafter plants. Concrete is a fast perishable good that must be produced just in time to meet customer demand requests, since it is not possible to store the concrete beforehand. The problem of production planning of concrete and dispatching concrete trucks is known in the literature as the ready-mix concrete dispatching problem (RMCDP), which is a generalization of both the vehicle routing and the parallel machine scheduling problems. In particular, this research deals with a real application on a major provider of concrete in Santiago, Chile, in which the scheduling of plants for production and the assignment of trucks for final dispatch of requests to the sites, are processes that have to be treated simultaneously. Although, customers, requests, resources and even the fleet, could vary from a day to another, the RMCDP we consider is a deterministic static problem solved daily by the company. We have implemented a GRASP heuristic algorithm to solve not only the static but also the dynamic versions of the problem. We have run many tests of the algorithm using historical data from the field provided by the company, obtaining promising results when compared to current practice. The final target is to implement a software in the company to make proper decisions of productions and assignment in order to minimize total costs, including mainly costs of delay, transportation and production.

Coauthors: Pablo A. Rey, Mauricio Cerda, Zdenko Koscina.

 

Dr. Gustavo Vulcano: Profesor plenario de Operaciones en la Escuela de Negocios de la Universidad Torcuato di Tella (UTDT), e investigador independiente del CONICET. Es también director del nuevo Master in Management (MiM) + Analytics en UTDT. El Prof. Vulcano se graduó como Licenciado en Ciencias de la Computación (UBA) en el año 1997, y posteriormente se doctoró en Decision, Risk and Operations en la Graduate School of Business, Columbia University (2003). Antes de incorporarse a di Tella, el Prof. Vulcano se desempeñó como profesor full-time en la Leonard N. Stern School of Business, New York University, durante el período 2002-2017, donde consiguió su promoción a Associate Professor with tenure en el año 2012. Actualmente se desempeña allí como profesor adjunto, enseñando dos cursos en el Master of Science in Business Analytics. Su trabajo de investigación incluye tópicos de revenue y pricing analytics, retail operations, y supply chain management. Sus papers han sido publicado en los más destacados journals internacionales de estas disciplinas, incluyendo Operations Research, Management Science, y Manufacturing and Service Operations Management. Actualmente se desempeña como co-editor de la nueva área de Revenue Management & Market Analytics en el journal Operations Research, y  como editor asociado del journal Management Science.  El Prof. Vulcano ha sido también Chair de la INFORMS Revenue Management and Pricing Section durante el período 2016-17, y ocupa ahora la posición de past-chair en el board. Asimismo, el Dr. Vulcano ha realizado trabajos de consultoría internacional en el área de servicios para Delta Airlines, Sabre Holdings y Movie Uruguay. En Argentina, ha sido consultor de la Fundación Pérez Companc, el Standard Bank (actualmente, ICBC), el grupo Flechabus, y Aerolíneas Argentinas.

Aplicaciones de grafos y programación entera para la personalización de ofertas en la industria retail.

Para un retailer, la implementación de promociones personalizadas es un medio para evitar el efecto negativo de las promociones masivas, como por ejemplo efectos de acumulación de stock por parte de los clientes aprovechando descuentos en productos que de todos modos comprarían. Las promociones personalizadas se apoyan en las preferencias heterogéneas de individuos para productos dentro de una categoría, y en sus diferentes reacciones a los precios descontados. En este trabajo consideramos el diseño de promociones personalizadas, y la predicción de las respuestas de los clientes a tales promociones sobre una categoría de productos. Finalmente, en base a esas predicciones, estudiamos el problema de optimizar el subconjunto de productos a promocionar durante la visita de un cliente a la tienda.

Los inputs requeridos por nuestra propuesta incluyen datos de panel con la historia de transacciones etiquetadas con el ID de cada cliente, información sobre el mix de productos disponible para una categoría de productos, y la identificación de productos que estaban en promoción en el momento de cada visita por parte de cada cliente.

Las preferencias de cada cliente se representan con un grafo acíclico dirigido (DAG). El DAG se construye en base a los datos históricos que revelan las preferencias de cada cliente. En el DAG, cada nodo i representa un producto, y el arco dirigido (i,j) representa la relación “el producto i es preferible sobre el producto j”.

Desde el punto de vista teórico, proveemos cotas computacionalmente tratables para calcular tanto la probabilidad de observar un DAG como las probabilidades de compra (que en general son problemas #P-hard). Luego, tomando la colección de DAGs representando la base de clientes, calibramos sobre sí variantes del modelo multinomial logit (MNL), junto a dos benchmarks que constituyen estados del arte en la materia (pero que no usan la estructura de DAGs). Finalmente, testeamos tanto nuestro modelo basado en DAGs como los dos benchmarks para el diseño de promociones personalizadas. Vía una formulación MIP, consideramos el problema de optimizar el subconjunto de productos a ofrecer en promoción para un cliente particular, dado un surtido de productos prefijado, con el objetivo de maximizar ingresos.

Los experimentos numéricos en datos de panel reales sobre 27 categorías de productos muestran que nuestra propuesta permite predicciones más certeras respecto a las compras de los clientes que los benchmarks testeados, con mejoras del orden de 10%. Finalmente, nuestro MIP para ofrecer promociones personalizadas permiten mejoras del 26% (en promedio) sobre los resultados que exhibe el dataset.

Trabajo conjunto con Srikanth Jagabathula y Dmitry Mitrofanov (New York University).