http://sedici.unlp.edu.ar:80/handle/10915/125192 Fri, 23 Oct 2026 19:44:40 GMT 2026-10-23T19:44:40Z http://sedici.unlp.edu.ar:80/handle/10915/125265 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) Feature selection is a useful machine learning technique aimed at reducing the dimensionality of the input space, discarding useless or redundant variables, in order to increase the performance and interpretability of models. The well-known Recursive Feature Elimination (RFE) algorithm provides good performance with moderate computational efforts, in particular for wide datasets. When using Support Vector Machines (SVM) for multiclass classification problems, the most typical strategy is to apply a simple One–Vs–One (OVO) strategy to produce a multiclass classifier starting from binary ones. In this work we introduce improved methods to produce the final ranking of features on multiclass problems with OVO–SVM, based on different combinations of the set of rankings produced by the diverse binary problems. We evaluated our new strategies using wide datasets from mass–spectrometry analysis and standard datasets from the UCI repository. In particular, we compared the new methods with the traditional average strategy. Our results suggest that one of our new methods outperforms the traditional scheme in most situations. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125265 2011-01-01T00:00:00Z Feature selection is a useful machine learning technique aimed at reducing the dimensionality of the input space, discarding useless or redundant variables, in order to increase the performance and interpretability of models. The well-known Recursive Feature Elimination (RFE) algorithm provides good performance with moderate computational efforts, in particular for wide datasets. When using Support Vector Machines (SVM) for multiclass classification problems, the most typical strategy is to apply a simple One–Vs–One (OVO) strategy to produce a multiclass classifier starting from binary ones. In this work we introduce improved methods to produce the final ranking of features on multiclass problems with OVO–SVM, based on different combinations of the set of rankings produced by the diverse binary problems. We evaluated our new strategies using wide datasets from mass–spectrometry analysis and standard datasets from the UCI repository. In particular, we compared the new methods with the traditional average strategy. Our results suggest that one of our new methods outperforms the traditional scheme in most situations. http://sedici.unlp.edu.ar:80/handle/10915/125264 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) Feature selection is a useful machine learning technique aimed at reducing the dimensionality of the input space, discarding useless or redundant variables, in order to increase the performance and interpretability of models. The well-known Recursive Feature Elimination (RFE) algorithm provides good performance with moderate computational efforts, in particular for wide datasets. When using Support Vector Machines (SVM) for multiclass classification problems, the most typical strategy is to apply a simple One–Vs–One (OVO) strategy to produce a multiclass classifier starting from binary ones. In this work we introduce improved methods to produce the final ranking of features on multiclass problems with OVO–SVM, based on different combinations of the set of rankings produced by the diverse binary problems. We evaluated our new strategies using wide datasets from mass–spectrometry analysis and standard datasets from the UCI repository. In particular, we compared the new methods with the traditional average strategy. Our results suggest that one of our new methods outperforms the traditional scheme in most situations. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125264 2011-01-01T00:00:00Z Feature selection is a useful machine learning technique aimed at reducing the dimensionality of the input space, discarding useless or redundant variables, in order to increase the performance and interpretability of models. The well-known Recursive Feature Elimination (RFE) algorithm provides good performance with moderate computational efforts, in particular for wide datasets. When using Support Vector Machines (SVM) for multiclass classification problems, the most typical strategy is to apply a simple One–Vs–One (OVO) strategy to produce a multiclass classifier starting from binary ones. In this work we introduce improved methods to produce the final ranking of features on multiclass problems with OVO–SVM, based on different combinations of the set of rankings produced by the diverse binary problems. We evaluated our new strategies using wide datasets from mass–spectrometry analysis and standard datasets from the UCI repository. In particular, we compared the new methods with the traditional average strategy. Our results suggest that one of our new methods outperforms the traditional scheme in most situations. http://sedici.unlp.edu.ar:80/handle/10915/125262 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) Este artículo presenta dos casos de estudio para el modelado de sistemas difusos basado en el Algoritmo Genético Compacto (CGA). Se estudian los casos de clasificación de la flor Iris de Fisher y la estimación de la función Sinc. Se realiza una comparación del enfoque basado en el CGA contra el enfoque basado en el Algoritmo Genético clásico (GA). Los resultados obtenidos de 100 experimentos para variaciones de los enfoques de CGA y GA, apuntan a que para los casos de estudio, el desempeño del enfoque basado en CGA es mejor dado que el error disminuye, y es más rápido con respecto al número de iteraciones necesarias para converger. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125262 2011-01-01T00:00:00Z Este artículo presenta dos casos de estudio para el modelado de sistemas difusos basado en el Algoritmo Genético Compacto (CGA). Se estudian los casos de clasificación de la flor Iris de Fisher y la estimación de la función Sinc. Se realiza una comparación del enfoque basado en el CGA contra el enfoque basado en el Algoritmo Genético clásico (GA). Los resultados obtenidos de 100 experimentos para variaciones de los enfoques de CGA y GA, apuntan a que para los casos de estudio, el desempeño del enfoque basado en CGA es mejor dado que el error disminuye, y es más rápido con respecto al número de iteraciones necesarias para converger. http://sedici.unlp.edu.ar:80/handle/10915/125261 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) Obtener un plan corpus a partir del cual aprender una biblioteca de planes que sirva de base de conocimiento para un sistema de reconocimiento de planes es una tarea difícil. En este trabajo se presenta un enfoque basado en planning para la generación automática de secuencias de acciones que llevan al cumplimiento de diferentes objetivos. El método propuesto permite adicionalmente la especificación de preferencias que se utilizan a la hora de elegir qué acción seguir en un punto determinado del plan de manera tal de poder modelar fácilmente diferentes perfiles de usuarios que utilicen el sistema. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125261 2011-01-01T00:00:00Z Obtener un plan corpus a partir del cual aprender una biblioteca de planes que sirva de base de conocimiento para un sistema de reconocimiento de planes es una tarea difícil. En este trabajo se presenta un enfoque basado en planning para la generación automática de secuencias de acciones que llevan al cumplimiento de diferentes objetivos. El método propuesto permite adicionalmente la especificación de preferencias que se utilizan a la hora de elegir qué acción seguir en un punto determinado del plan de manera tal de poder modelar fácilmente diferentes perfiles de usuarios que utilicen el sistema. http://sedici.unlp.edu.ar:80/handle/10915/125258 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) The use of wireless sensor networks for information discovery and monitoring of continuous physical fields has emerged as a novel and efficient solution. To this end, a query message is routed through the network to fetch data from sensor nodes and report it back to a sink node. As several applications only require a limited subset of the available data in the network, this query could be ideally routed to fetch only relevant data. In this way, much energy due to message exchange among nodes could be saved. In this paper, we consider the application of computational intelligence on nodes to implement a parallel adaptive simulated annealing (PASA) mechanism able to direct queries to relevant nodes. Besides, a reinforcement learning algorithm is proposed to adapt progressively the query process to the characteristics of the network, limiting the routing space to areas with useful data. Finally, the relevant data collection mechanism is also discussed to illustrate the complete process. We show by extensive simulations that the routing cost can be reduced by approximately 60% over flooding with an error less than 5%. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125258 2011-01-01T00:00:00Z The use of wireless sensor networks for information discovery and monitoring of continuous physical fields has emerged as a novel and efficient solution. To this end, a query message is routed through the network to fetch data from sensor nodes and report it back to a sink node. As several applications only require a limited subset of the available data in the network, this query could be ideally routed to fetch only relevant data. In this way, much energy due to message exchange among nodes could be saved. In this paper, we consider the application of computational intelligence on nodes to implement a parallel adaptive simulated annealing (PASA) mechanism able to direct queries to relevant nodes. Besides, a reinforcement learning algorithm is proposed to adapt progressively the query process to the characteristics of the network, limiting the routing space to areas with useful data. Finally, the relevant data collection mechanism is also discussed to illustrate the complete process. We show by extensive simulations that the routing cost can be reduced by approximately 60% over flooding with an error less than 5%. http://sedici.unlp.edu.ar:80/handle/10915/125255 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) Se diseño un modelo de red neuronal artificial para la predicción al día siguiente del máximo diario de PM10, (material particulado de menos de 10 micrometros de diámetro), el cual se construye de manera dinámica mediante la formación de clusters para la clasificación de patrones y evoluciona a través de los datos que recibe automáticamente y en tiempo real. Se generó una matriz de distancias a partir de los patrones de entrada para seleccionar el radio óptimo de clasificación. El modelo fue validado mediante la aplicación de datos históricos de variables meteorológicas y de PM10 registrados en Villahermosa, Tabasco, México de 2007 a 2009. Los experimentos realizados permitieron identificar las variables relevantes del modelo y se contemplaron datos normalizados y no-normalizados. Los mejores resultados del modelo se obtuvieron usando promedios móviles y valores máximos y mínimos de PM10 no normalizados como variables de entrada así como radios cercanos al valor mínimo calculado en la matriz de distancias. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125255 2011-01-01T00:00:00Z Se diseño un modelo de red neuronal artificial para la predicción al día siguiente del máximo diario de PM10, (material particulado de menos de 10 micrometros de diámetro), el cual se construye de manera dinámica mediante la formación de clusters para la clasificación de patrones y evoluciona a través de los datos que recibe automáticamente y en tiempo real. Se generó una matriz de distancias a partir de los patrones de entrada para seleccionar el radio óptimo de clasificación. El modelo fue validado mediante la aplicación de datos históricos de variables meteorológicas y de PM10 registrados en Villahermosa, Tabasco, México de 2007 a 2009. Los experimentos realizados permitieron identificar las variables relevantes del modelo y se contemplaron datos normalizados y no-normalizados. Los mejores resultados del modelo se obtuvieron usando promedios móviles y valores máximos y mínimos de PM10 no normalizados como variables de entrada así como radios cercanos al valor mínimo calculado en la matriz de distancias. http://sedici.unlp.edu.ar:80/handle/10915/125252 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) Dimension reduction is often included in pattern recognizers based on hidden Markov models to lower the size of the models to estimate. Commonly used methods are heuristic in nature and do not take care of information retention after projection. In this paper, we present a new method based on the approach of suficient dimension reductions. It explicitly accounts for all the discriminative information available in the original features, while using a minimum number of linear combinations of them. We review the underlying theory and present an algorithm for the practical implementation of the proposed method. On the experimental side, we use simulations to illustrate its advantages over widely-used existing alternatives. In particular, we show that it performs as good as existing techniques when data is optimal according to the assumptions of those techniques, but signi cantly better for heteroscedastic data with no special structure on the covariance matrix. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125252 2011-01-01T00:00:00Z Dimension reduction is often included in pattern recognizers based on hidden Markov models to lower the size of the models to estimate. Commonly used methods are heuristic in nature and do not take care of information retention after projection. In this paper, we present a new method based on the approach of suficient dimension reductions. It explicitly accounts for all the discriminative information available in the original features, while using a minimum number of linear combinations of them. We review the underlying theory and present an algorithm for the practical implementation of the proposed method. On the experimental side, we use simulations to illustrate its advantages over widely-used existing alternatives. In particular, we show that it performs as good as existing techniques when data is optimal according to the assumptions of those techniques, but signi cantly better for heteroscedastic data with no special structure on the covariance matrix. http://sedici.unlp.edu.ar:80/handle/10915/125251 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) In this work, we present an instantiation of our framework for Hierarchical Distance-based Conceptual Clustering (HDCC) using sequences, a particular kind of structured data. We analyze the relationship between distances and generalization operators for sequences in the context of HDCC. HDCC is a general approach to conceptual clustering that extends the traditional algorithm for hierarchical clustering by producing conceptual generalizations of the discovered clusters. Since the approach is general, it allows combining the flexibility of changing distances for different data types at the same time that we take advantage of the interpretability offered by the obtained concepts, which is central for descriptive data mining tasks. We propose here different generalization operators for sequences and analyze how they work together with the edit and linkage distances in HDCC. This analysis is carried out based on three different properties for generalization operators and three different levels of agreement between the clustering hierarchy obtained from the linkage distance and the hierarchy obtained by using generalization operators. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125251 2011-01-01T00:00:00Z In this work, we present an instantiation of our framework for Hierarchical Distance-based Conceptual Clustering (HDCC) using sequences, a particular kind of structured data. We analyze the relationship between distances and generalization operators for sequences in the context of HDCC. HDCC is a general approach to conceptual clustering that extends the traditional algorithm for hierarchical clustering by producing conceptual generalizations of the discovered clusters. Since the approach is general, it allows combining the flexibility of changing distances for different data types at the same time that we take advantage of the interpretability offered by the obtained concepts, which is central for descriptive data mining tasks. We propose here different generalization operators for sequences and analyze how they work together with the edit and linkage distances in HDCC. This analysis is carried out based on three different properties for generalization operators and three different levels of agreement between the clustering hierarchy obtained from the linkage distance and the hierarchy obtained by using generalization operators. http://sedici.unlp.edu.ar:80/handle/10915/125249 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) La inclusión de las Nuevas Tecnologías de la Información y Comunicación (NTIC), provocó cambios en diferentes áreas universitarias, una de esas áreas son las bibliotecas. La digitalización de documentos llevó a replantear las herramientas utilizadas para la búsqueda de información, dando surgimiento a las hoy conocidas Bibliotecas Digitales. Generalmente, debido a la cantidad de información que cuentan estas bibliotecas, realizar una consulta resulta una tarea tediosa que consume una considerable cantidad de tiempo. Esta problemática podría solucionarse con la incorporación de agentes inteligentes que mediante una red bayesiana filtren y seleccionen la información, acorde con lo solicitado por el usuario, pero teniendo en cuenta también sus gustos y preferencias. El presente trabajo se enfoca en la creación de un prototipo de Biblioteca Digital Inteligente destinada a investigadores, profesores y estudiantes de la Facultad de Ciencias Forestales de la Universidad Nacional de Santiago del Estero, que mediante la incorporación de agentes, será capaz de administrar modelos de usuarios para brindar servicios personalizados de manera eficaz y sencilla. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125249 2011-01-01T00:00:00Z La inclusión de las Nuevas Tecnologías de la Información y Comunicación (NTIC), provocó cambios en diferentes áreas universitarias, una de esas áreas son las bibliotecas. La digitalización de documentos llevó a replantear las herramientas utilizadas para la búsqueda de información, dando surgimiento a las hoy conocidas Bibliotecas Digitales. Generalmente, debido a la cantidad de información que cuentan estas bibliotecas, realizar una consulta resulta una tarea tediosa que consume una considerable cantidad de tiempo. Esta problemática podría solucionarse con la incorporación de agentes inteligentes que mediante una red bayesiana filtren y seleccionen la información, acorde con lo solicitado por el usuario, pero teniendo en cuenta también sus gustos y preferencias. El presente trabajo se enfoca en la creación de un prototipo de Biblioteca Digital Inteligente destinada a investigadores, profesores y estudiantes de la Facultad de Ciencias Forestales de la Universidad Nacional de Santiago del Estero, que mediante la incorporación de agentes, será capaz de administrar modelos de usuarios para brindar servicios personalizados de manera eficaz y sencilla. http://sedici.unlp.edu.ar:80/handle/10915/125248 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) We introduce in this work a set of strategies for improving the piecing-together step in local-to-global Markov networks structure learning algorithms. For Markov networks, Local-to-global algorithms decompose the problem of learning a complete independence structure with n variables into n independent Markov blanket learning problems. On a second step these algorithms piece-together all the learned Markov blankets into a global structure using an \OR rule". Insucient data may result in incorrect learning of Markov blankets, with con in their decision on edge inclusion when, for two variables X and Y , X is in the blanket of Y , but Y is not in the blanket of X. In such cases the \OR rule" always decides to add the edge, making mistakes when such edge does not exist. Our contribution is alternative strategies. The first alternative is based on the \AND rule" which proposes to add an edge between two variables X and Y to the global structure if they mutually belong to its respective Markov blankets. The other alternative rule is based on the probability of the edges and aims to solve an inconsistency by comparing the probability of edge existence with the probability of edge absence, and taking the more probable for deciding to add or remove such edge. At the end of the work, we show that inconsistencies are an important source of errors in these algorithms, and demonstrate empirically interesting improvements in the quality of learned structures, using this new piecing-together alternative instead of the basic \OR rule". Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125248 2011-01-01T00:00:00Z We introduce in this work a set of strategies for improving the piecing-together step in local-to-global Markov networks structure learning algorithms. For Markov networks, Local-to-global algorithms decompose the problem of learning a complete independence structure with n variables into n independent Markov blanket learning problems. On a second step these algorithms piece-together all the learned Markov blankets into a global structure using an \OR rule". Insucient data may result in incorrect learning of Markov blankets, with con in their decision on edge inclusion when, for two variables X and Y , X is in the blanket of Y , but Y is not in the blanket of X. In such cases the \OR rule" always decides to add the edge, making mistakes when such edge does not exist. Our contribution is alternative strategies. The first alternative is based on the \AND rule" which proposes to add an edge between two variables X and Y to the global structure if they mutually belong to its respective Markov blankets. The other alternative rule is based on the probability of the edges and aims to solve an inconsistency by comparing the probability of edge existence with the probability of edge absence, and taking the more probable for deciding to add or remove such edge. At the end of the work, we show that inconsistencies are an important source of errors in these algorithms, and demonstrate empirically interesting improvements in the quality of learned structures, using this new piecing-together alternative instead of the basic \OR rule". http://sedici.unlp.edu.ar:80/handle/10915/125247 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) Organizar a los estudiantes en grupos para que trabajen juntos no garantiza que alcancen un aprendizaje colaborativo efectivo. Las conductas que manifiesta cada uno de los integrantes, es decir los roles que se asuman, son fundamentales para alcanzar con éxito los objetivos de enseñanza y de aprendizaje. Por ello, en este trabajo se presenta un proyecto que propone un modelo multiagente que monitorea la participación de los estudiantes dentro del grupo, reconoce los roles que ellos desenvuelven al trabajar colaborativamente, construye automáticamente sus perfiles de usuario, diagnostica el estado de la colaboración considerando el equilibrio de roles como situación ideal, y propone acciones correctivas cuando el comportamiento del grupo se aleja de dicho equilibrio. El modelo a desarrollar será implementado en un entorno de aprendizaje a distancia, y para validar su funcionamiento será utilizado por grupos de estudiantes reales durante sesiones de trabajo colaborativo. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125247 2011-01-01T00:00:00Z Organizar a los estudiantes en grupos para que trabajen juntos no garantiza que alcancen un aprendizaje colaborativo efectivo. Las conductas que manifiesta cada uno de los integrantes, es decir los roles que se asuman, son fundamentales para alcanzar con éxito los objetivos de enseñanza y de aprendizaje. Por ello, en este trabajo se presenta un proyecto que propone un modelo multiagente que monitorea la participación de los estudiantes dentro del grupo, reconoce los roles que ellos desenvuelven al trabajar colaborativamente, construye automáticamente sus perfiles de usuario, diagnostica el estado de la colaboración considerando el equilibrio de roles como situación ideal, y propone acciones correctivas cuando el comportamiento del grupo se aleja de dicho equilibrio. El modelo a desarrollar será implementado en un entorno de aprendizaje a distancia, y para validar su funcionamiento será utilizado por grupos de estudiantes reales durante sesiones de trabajo colaborativo. http://sedici.unlp.edu.ar:80/handle/10915/125246 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) La omisión de acentos ortograáficos es un error tipográfi co muy frecuente en el idioma español; su restauración automática consiste en la inserción de acentos omitidos en los lugares que son necesarios. Los adverbios interrogativos son un caso especialmente di ficultoso de este problema, ya que en muchas ocasiones no existen marcas claras que indiquen su presencia. Este trabajo presenta dos técnicas de aprendizaje automático, Conditional Random Fields (CRF) y Support Vector Ma- chines (SVM), aplicadas a la resolución del problema de la restauración automática de acentos ortográ cos para el caso especifí co de los adverbios interrogativos. Se obtuvieron buenos resultados con ambas técnicas, siendo sensiblemente superior el resultado obtenido utilizando un clasificador basado en CRF, y que utiliza como atributos los tokens que más comúnmente preceden y siguen a los adverbios interrogativos. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125246 2011-01-01T00:00:00Z La omisión de acentos ortograáficos es un error tipográfi co muy frecuente en el idioma español; su restauración automática consiste en la inserción de acentos omitidos en los lugares que son necesarios. Los adverbios interrogativos son un caso especialmente di ficultoso de este problema, ya que en muchas ocasiones no existen marcas claras que indiquen su presencia. Este trabajo presenta dos técnicas de aprendizaje automático, Conditional Random Fields (CRF) y Support Vector Ma- chines (SVM), aplicadas a la resolución del problema de la restauración automática de acentos ortográ cos para el caso especifí co de los adverbios interrogativos. Se obtuvieron buenos resultados con ambas técnicas, siendo sensiblemente superior el resultado obtenido utilizando un clasificador basado en CRF, y que utiliza como atributos los tokens que más comúnmente preceden y siguen a los adverbios interrogativos. http://sedici.unlp.edu.ar:80/handle/10915/125243 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) This work presents a quantum search algorithm to solve the 3-SAT problem. An improvement over one of the best known classical algorithms for this problem is proposed, replacing the local search with a quantum search algorithm. The performance of the improved algorithm is assessed by simulating it using parallel programming techniques with shared memory. The experimental analysis demonstrate that the parallel simulation of the algorithm takes advantage of the available computing resources to improve over the eficiency of the sequential version, thus allowing to perform realistic simulations in reduced execution times. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125243 2011-01-01T00:00:00Z This work presents a quantum search algorithm to solve the 3-SAT problem. An improvement over one of the best known classical algorithms for this problem is proposed, replacing the local search with a quantum search algorithm. The performance of the improved algorithm is assessed by simulating it using parallel programming techniques with shared memory. The experimental analysis demonstrate that the parallel simulation of the algorithm takes advantage of the available computing resources to improve over the eficiency of the sequential version, thus allowing to perform realistic simulations in reduced execution times. http://sedici.unlp.edu.ar:80/handle/10915/125241 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) Los líderes de equipo promueven la re exión, coordinan al grupo, y explotan los puntos fuertes del equipo para cumplir sus objetivos. De esta manera, el liderazgo se convierte en un factor determinante para lograr el éxito de cualquier equipo. En este trabajo se propone un modelo para la detección automática de líderes de equipo a partir del análisis de las interacciones de los usuarios trabajando colaborativamente a través de un sistema para trabajo en grupo. Particularmente, se propone aplicar técnicas de aprendizaje automático sobre los datos observados del desempeño de los usuarios trabajando en conjunto para determinar si sus habilidades y características le permiten desempeñar el rol de líder dentro de un equipo. Para alcanzar este objetivo se analizó y evaluó el desempeño de diferentes técnicas de clasi cación. Los resultados obtenidos demuestran que la técnica propuesta permite detectar los líderes con una buena precisión y podría utilizarse exitosamente en la formación de equipos virtuales. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125241 2011-01-01T00:00:00Z Los líderes de equipo promueven la re exión, coordinan al grupo, y explotan los puntos fuertes del equipo para cumplir sus objetivos. De esta manera, el liderazgo se convierte en un factor determinante para lograr el éxito de cualquier equipo. En este trabajo se propone un modelo para la detección automática de líderes de equipo a partir del análisis de las interacciones de los usuarios trabajando colaborativamente a través de un sistema para trabajo en grupo. Particularmente, se propone aplicar técnicas de aprendizaje automático sobre los datos observados del desempeño de los usuarios trabajando en conjunto para determinar si sus habilidades y características le permiten desempeñar el rol de líder dentro de un equipo. Para alcanzar este objetivo se analizó y evaluó el desempeño de diferentes técnicas de clasi cación. Los resultados obtenidos demuestran que la técnica propuesta permite detectar los líderes con una buena precisión y podría utilizarse exitosamente en la formación de equipos virtuales. http://sedici.unlp.edu.ar:80/handle/10915/125240 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) Web Directories consist of large collections of links to websites, arranged by topic in different categories. The structure of Web Directories is typically not flat, since there are hierarchical and non-hierarchical relationships among topics. As a consequence, websites classified under certain topic may be relevant to other topics. While some of these relevance relations are explicit, most of them must be discovered by an analysis of the structure of these directories. This paper proposes a family of models of relevance propagation in Web Directories. An efficient computational framework for inferring implicit relevance relations is described. The framework presented here combines the use of matrices to represent relevance relations and the application of boolean operations on these matrices to infer implicit relations. Nine different models were computed for a portion of the Open Directory Project graph consisting of more than half a million nodes and approximately 1.5 million edges of different types. The models were compared by performing both a quantitative and qualitative analysis on them. It is found that some general difficulties rule out the possibility of defining flawless models of relevance propagation that only take into account structural features of Web Directories. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125240 2011-01-01T00:00:00Z Web Directories consist of large collections of links to websites, arranged by topic in different categories. The structure of Web Directories is typically not flat, since there are hierarchical and non-hierarchical relationships among topics. As a consequence, websites classified under certain topic may be relevant to other topics. While some of these relevance relations are explicit, most of them must be discovered by an analysis of the structure of these directories. This paper proposes a family of models of relevance propagation in Web Directories. An efficient computational framework for inferring implicit relevance relations is described. The framework presented here combines the use of matrices to represent relevance relations and the application of boolean operations on these matrices to infer implicit relations. Nine different models were computed for a portion of the Open Directory Project graph consisting of more than half a million nodes and approximately 1.5 million edges of different types. The models were compared by performing both a quantitative and qualitative analysis on them. It is found that some general difficulties rule out the possibility of defining flawless models of relevance propagation that only take into account structural features of Web Directories. http://sedici.unlp.edu.ar:80/handle/10915/125239 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) In the context of Mobile Robotics, the e cient resolution of the Path Planning problem is a key task. The model of the environment and the search algorithm are basic issues in the resolution of the problem. This paper highlights the main features of Path Planning proposal for mobile robots in static environments. In our proposal, the path planning is based on Voronoi diagrams, where obstacles in the environment are considered as the generating points of the diagram, and a genetic algorithm is used to nd a path without collisions from the robot initial to target position. This work combines some ideas presented by Roque and Doering, who use Voronoi diagrams for modelling the environment, and other ideas presented by Zhang et al. who adopt a genetic algorithm for computing paths on a regular grid based environment, considering certain quality attributes. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125239 2011-01-01T00:00:00Z In the context of Mobile Robotics, the e cient resolution of the Path Planning problem is a key task. The model of the environment and the search algorithm are basic issues in the resolution of the problem. This paper highlights the main features of Path Planning proposal for mobile robots in static environments. In our proposal, the path planning is based on Voronoi diagrams, where obstacles in the environment are considered as the generating points of the diagram, and a genetic algorithm is used to nd a path without collisions from the robot initial to target position. This work combines some ideas presented by Roque and Doering, who use Voronoi diagrams for modelling the environment, and other ideas presented by Zhang et al. who adopt a genetic algorithm for computing paths on a regular grid based environment, considering certain quality attributes. http://sedici.unlp.edu.ar:80/handle/10915/125237 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) El objetivo de este trabajo es predecir la trayectoria de un móvil aéreo cuando por diferentes razones deja de recibirse la información GPS sobre su posición actual. Específicamente, se desarrolla un filtro no lineal basado en redes neuronales feedforward para la estimación de dicha trayectoria. A continuación, se analiza el desempeño en dos casos: utilizando un patrón de entradas de tamaño fijo y otro de dimensión variable. Este trabajo muestra los resultados pertenecientes al plano vertical siendo este análogo para latitud y longitud. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125237 2011-01-01T00:00:00Z El objetivo de este trabajo es predecir la trayectoria de un móvil aéreo cuando por diferentes razones deja de recibirse la información GPS sobre su posición actual. Específicamente, se desarrolla un filtro no lineal basado en redes neuronales feedforward para la estimación de dicha trayectoria. A continuación, se analiza el desempeño en dos casos: utilizando un patrón de entradas de tamaño fijo y otro de dimensión variable. Este trabajo muestra los resultados pertenecientes al plano vertical siendo este análogo para latitud y longitud. http://sedici.unlp.edu.ar:80/handle/10915/125236 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) Multiclass problems are usually of high technological value, but many classification methods are binary in origin. In the last years, several improved solutions based on the combination of simple classifiers were introduced. An interesting solution is based on creating a hierarchy of sub-problems by clustering prototypes of each one of the classes; there- fore the solution is heavily influenced by the label’s information. In this work we analyze a new strategy to solve multiclass problems that makes more use of spatial information than other methods. We construct a hier- archy of subproblems, but opposite to previous developments, based only on spatial information and not using a single prototype for each class. We evaluate the use of different clustering methods (either agglomera- tive or divisive) for this task and also the use two different classifiers (linear SVM and FDA–GenRidge) for each sub-problem (if needed, be- cause in several cases the clustering method directly gives a subset with samples of a single class). We compare the new method with several pre- vious approaches, finding promising results. The good performance of our approach is virtually independent of the classifier coupled to it, which suggest that it success is primarily related to the use of an appropriate clustering strategy. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125236 2011-01-01T00:00:00Z Multiclass problems are usually of high technological value, but many classification methods are binary in origin. In the last years, several improved solutions based on the combination of simple classifiers were introduced. An interesting solution is based on creating a hierarchy of sub-problems by clustering prototypes of each one of the classes; there- fore the solution is heavily influenced by the label’s information. In this work we analyze a new strategy to solve multiclass problems that makes more use of spatial information than other methods. We construct a hier- archy of subproblems, but opposite to previous developments, based only on spatial information and not using a single prototype for each class. We evaluate the use of different clustering methods (either agglomera- tive or divisive) for this task and also the use two different classifiers (linear SVM and FDA–GenRidge) for each sub-problem (if needed, be- cause in several cases the clustering method directly gives a subset with samples of a single class). We compare the new method with several pre- vious approaches, finding promising results. The good performance of our approach is virtually independent of the classifier coupled to it, which suggest that it success is primarily related to the use of an appropriate clustering strategy. http://sedici.unlp.edu.ar:80/handle/10915/125233 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) La clasificación de documentos de texto es una aplicación de la minería de textos que pretende extraer información de texto no estructurado. Su interés se justifica porque se estima que entre el 80% y el 90% de los datos de las organizaciones son no estructurados. Por otro lado, la búsqueda semántica permite al usuario especificar en una consulta no solamente términos que deben aparecer en el documento, sino conceptos y relaciones, que pueden detectarse mediante el análisis de texto. El objetivo de este trabajo es implementar un buscador semántico que aproveche el resultado de algoritmos de aprendizaje automático supervisado y semi-supervisado para la categorización o clasificación de documentos. El dominio de aplicación es un corpus de más de 8000 documentos que contienen nueve años de resoluciones rectorales de la Universidad Católica de Salta en distintos formatos. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125233 2011-01-01T00:00:00Z La clasificación de documentos de texto es una aplicación de la minería de textos que pretende extraer información de texto no estructurado. Su interés se justifica porque se estima que entre el 80% y el 90% de los datos de las organizaciones son no estructurados. Por otro lado, la búsqueda semántica permite al usuario especificar en una consulta no solamente términos que deben aparecer en el documento, sino conceptos y relaciones, que pueden detectarse mediante el análisis de texto. El objetivo de este trabajo es implementar un buscador semántico que aproveche el resultado de algoritmos de aprendizaje automático supervisado y semi-supervisado para la categorización o clasificación de documentos. El dominio de aplicación es un corpus de más de 8000 documentos que contienen nueve años de resoluciones rectorales de la Universidad Católica de Salta en distintos formatos. http://sedici.unlp.edu.ar:80/handle/10915/125228 Objeto de conferencia XII Argentine Symposium on Artificial Intelligence (ASAI 2011) (XL JAIIO, Córdoba, 29 y 30 de agosto de 2011) Textual conversations on the Internet, such us chat rooms or instant messaging services, have become an excellent source of data for semantic analysis. In particular, potential user interests or user-related topics could be extracted from these conversations for personalization purposes. In this work, we present a novel method for topic detection from chat logs. First, we de ned the generic structure of the process. Then, a variety of text-mining techniques was evaluated in each step of the process. Stemming, synonyms, POS tagging and named entities recognition are examples of these techniques. Encouraging experimental results from a comparative evaluation procedure, allow us to determine the most suitable combination of techniques for the problem. Sat, 01 Jan 2011 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/125228 2011-01-01T00:00:00Z Textual conversations on the Internet, such us chat rooms or instant messaging services, have become an excellent source of data for semantic analysis. In particular, potential user interests or user-related topics could be extracted from these conversations for personalization purposes. In this work, we present a novel method for topic detection from chat logs. First, we de ned the generic structure of the process. Then, a variety of text-mining techniques was evaluated in each step of the process. Stemming, synonyms, POS tagging and named entities recognition are examples of these techniques. Encouraging experimental results from a comparative evaluation procedure, allow us to determine the most suitable combination of techniques for the problem.