Detección de género mediante discriminadores lineales: Reconocimiento de género usando técnicas lineales: El reconocimiento del género de una persona a partir de una im agen de su cara es uno de los problemas de clasificación demográfica estudiados en los ultimos años. Según Juan Bekios-Calfa, José Buenaposada y Luis Baumela, en su paper "Revisiting Linear Discriminant Techniques in Gender Recognition", las soluciones posibles para este problema pueden agruparse en enfoques basados en la apariencia y enfoques basados en características. El enfoque basado en la apariencia consisten en aplicar técnicas de recorte y redimension de la imagen y métodos de normalización de la iluminación de la misma, y utilizar toda la cara resultante como atributos de clasificación. Por otro lado, el enfoques basado en características consiste en la extracción de un conjunto de característisticas relevantes y discriminantes de cada imagen, las cuales son los atributos utilizados para realizar la clasificación; de esta manera se evita trabajar con las im ́agenes originales que pueden presentar problemas debido a su alta dimensionalidad, entre otras cosas. Sin embargo, uno no conoce a priori cuáles son esas características relevantes y discriminantes; para ello se utilizan diferentes ténicas, entre las cuales se encuentran LDA (Linear Discriminant Analysis), PCA (Principal Components Analysis) e ICA (Independent Components Analysis). En este taller se abordará el problema del reconocimiento del género de una persona mediante un enfoque basado en la extracción de características que nos permitan realizar dicha discriminación, combinando LDA y PCA para determinar las mismas. El algoritmo a trabajar corresponde a uno de los presentados en el paper arriba mencionado. Si bien el taller es autocontenido, se recomienda mirar algo de Matlab, versión 2012a en adelante y refrescar conocimientos de manejo de matrices, autovectores, autovalores, matriz de covarianza.

Statistical Information Theory: Statistics and Information Theory in Remote Sensing with SAR: Statistics, either implicitly or explicitly, plays a prominent role across several branches of Remote Sensing (RS). This is mostly due to the fact that RS deals with often incomplete and mostly imprecise data. But Statistics aspires to more than being a mere tool for circumventing those unavoidable observational limitations. Statistics is able to provide a complete framework for tackling many relevant RS problems, from a sound mathematical description to tractable computational solutions. This wealth of knowledge is of particular importance when dealing with Synthetic Aperture Radar - SAR images. Thins kind of imaging produces data with a noise-like pattern, called speckle, which can be well described as a non-Gaussian non-additive contamination to the underlying desired information. Tools firmly grounded in a statistical approach are among the best suited for SAR image processing and analysis. In this talk we present a unified framework for a diversity of problems involving SAR imagery (despecking filters, classification, segmentation, change detection and edge identification). Using Information-Theoretic tools within a Statistical framework, we show that all these seemingly different problems can be posed and solved as a single one: testing the hypothesis that two or more samples are outcomes of the same distribution. Although the examples are instantiated for SAR, the framework is general enough to encompass a large variety of problems, including other models and types of data.

Dr Frery received the Engineer's Degree in Electronics and Electricity from the Facultad de Ingeniería, Universidad de Mendoza, Argentina in 1985. His MSc degree is in Applied Mathematics (Statistics) from the Instituto de Matemática Pura e Aplicada (IMPA), Rio de Janeiro, RJ, Brazil, obtained in 1990. His PhD degree is in Applied Computing from the Instituto Nacional de Pesquisas Espaciais (INPE), São José dos Campos, SP, Brazil, obtained in 1993. During both his MSc and PhD he worked with statistical models for image simulation, processing, and analysis: Markov random fields and models for synthetic aperture radar (SAR) images. His current research interests, besides image processing and analysis, include computational statistics and modelling of complex systems.

Support Vector Machines for Pattern Recognition of Signals: The Tutorial is designed for students without previous experience in Support Vector Machines (SVM). The aim of the tutorial is to introduce the SVM and present its comparative advantages when they are used as classifiers. The Tutorial is divided into two stages: the first stage is an introduction to pattern recognition and to the basic concepts of SVM. The second stage presents a work in which these tools were used for pattern recognition of seismic and biomedical signals.

Millaray Curilem received the B.E. degree in electrical engineering from the Instituto Superior Politécnico, José Antonio Echeverría, Cuba, in 1991, and the Dr. degree in Electrical Engineering from the Universidade Federal de Santa Catarina, Brazil, in 2002. In 1994, she joined the Department of Electrical Engineering, of the Universidad de La Frontera, at Temuco, Chile, as a Lecturer, and in 2009 she became Associate Professor of this University. She performed a post-doctoral research in the Department of Informatics at the Universidad de Santiago de Chile (2007). Her current research interests include Computational Intelligence applied to pattern recognition of seismic signals from volcanoes, biomedical engineering and education. Dr. Curilem is a Fellow Member of the Computational Intelligence Society of the IEEE. She participated in the Organizing Committee of the Iberoamerican Congress on Pattern Recognition (CIARP 2011) and the Chilean Workshop on Pattern Recognition (CWPR 2010-2011). Dr. Curilem has many ISI, Scopus and SciELO publications and she participates in several national and institutional research projects as main or co-researcher.

Intel Xeon Phi Programming: In recent years, the use of compute-intensive coprocessors has been very studied in the eld of Parallel Computing, used (among other things) to accelerate sequential processes. Up to now, the NVIDIA GPU (Graphics Processing Units) has been the most popular coprocessor. It applies a many-core architecture (massive multi-core architecture), i.e. it consists of a huge quantity of cores compared to a conventional multi-core architecture. Recently, Intel has released a coprocessor called Xeon Phi, based on the Intel MIC architecture, which is composed of up to 61 cores, and achieves a performance of up to 1208 GFLOP/s (billions of floating point operations per second) with a double precision floating point. This tutorial will show an overview of the architecture used by the Intel Xeon Phi and its programming model. We will compile and analyze different examples using the Xeon Phi on lab, and we also will describe the key factors to exploit efficiently this coprocessor.

Dr Barrientos received the PhD in Computer Science from the Complutense University of Madrid (Spain). He received two MSc in Computer Sciences, the first from the University of Chile (Chile) and the second from the Complutense University of Madrid (Spain), and the Computer Engineering from the University of Magallanes (Chile). Areas of interest: Parallel Computing and Information Retrieval. Currently, he is a professor at the Catholic University of Maule where he leads the research area of Parallel Computing