http://sedici.unlp.edu.ar:80/handle/10915/140166 Sun, 15 Nov 2026 05:16:17 GMT 2026-11-15T05:16:17Z http://sedici.unlp.edu.ar:80/handle/10915/140249 Objeto de conferencia XXII Simposio Argentino de Inteligencia artificial (ASSAI 2021) - JAIIO 50 (Modalidad virtual) Controller synthesis has been used in recent research to generate, from user specifications, correctby-construction motion and task plans for mobile robots. These plans are implemented by hybrid control architectures allowing the robot to move and interact with complex environments. However, the different abstraction levels involved together with the many components that must be developed for a fully functional end-to-end system requires numerous design and methodology decisions, specially when the robot strongly interacts with its environment. In this paper, we present an end-to-end system design and implementation for a mobile robot targeted at warehouse applications, that uses a proof-of- concept approach in a simulated environment as a key step in the design process. We demonstrate its capabilities in product rearrangement both in simulated and real world scenarios. Fri, 01 Jan 2021 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/140249 2021-01-01T00:00:00Z Controller synthesis has been used in recent research to generate, from user specifications, correctby-construction motion and task plans for mobile robots. These plans are implemented by hybrid control architectures allowing the robot to move and interact with complex environments. However, the different abstraction levels involved together with the many components that must be developed for a fully functional end-to-end system requires numerous design and methodology decisions, specially when the robot strongly interacts with its environment. In this paper, we present an end-to-end system design and implementation for a mobile robot targeted at warehouse applications, that uses a proof-of- concept approach in a simulated environment as a key step in the design process. We demonstrate its capabilities in product rearrangement both in simulated and real world scenarios. http://sedici.unlp.edu.ar:80/handle/10915/140247 Objeto de conferencia XXII Simposio Argentino de Inteligencia artificial (ASSAI 2021) - JAIIO 50 (Modalidad virtual) This work presents Clean-Dirty Containers in Montevideo (CDCM), a novel dataset for detection and classi cation of residue con-tainers. Images were collected from several sources, including Google Street View, Social Networks and smarthpone taken photos. The dataset is publicly available under a Creative Commons License. Fri, 01 Jan 2021 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/140247 2021-01-01T00:00:00Z This work presents Clean-Dirty Containers in Montevideo (CDCM), a novel dataset for detection and classi cation of residue con-tainers. Images were collected from several sources, including Google Street View, Social Networks and smarthpone taken photos. The dataset is publicly available under a Creative Commons License. http://sedici.unlp.edu.ar:80/handle/10915/140245 Objeto de conferencia XXII Simposio Argentino de Inteligencia artificial (ASSAI 2021) - JAIIO 50 (Modalidad virtual) Autonomous car driving has gained popularity in recent years both in industrial and academic environments, leading to numerous commercial and research platforms being constructed for development and testing. Moreover, due to advantages at design time and for testing purposes, several simulation environments have been proposed to simulate the complex dynamics and real-world problems the driving algorithms will face. However, most platforms and simulation environments are not suitable in costs and complexity for research groups beginning to work in the area or for educational purposes. In this work we propose a lowcost 3D-printed hardware 1:10 scale car that allows for development and testing of near SAE level 2 driving automation in software-in-the-loop realistic simulations together with real world scenarios. We demonstrate this by running a lane-centering and a simplified version of adaptive cruise control algorithm in a simulated and real highway-inspired track. Fri, 01 Jan 2021 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/140245 2021-01-01T00:00:00Z Autonomous car driving has gained popularity in recent years both in industrial and academic environments, leading to numerous commercial and research platforms being constructed for development and testing. Moreover, due to advantages at design time and for testing purposes, several simulation environments have been proposed to simulate the complex dynamics and real-world problems the driving algorithms will face. However, most platforms and simulation environments are not suitable in costs and complexity for research groups beginning to work in the area or for educational purposes. In this work we propose a lowcost 3D-printed hardware 1:10 scale car that allows for development and testing of near SAE level 2 driving automation in software-in-the-loop realistic simulations together with real world scenarios. We demonstrate this by running a lane-centering and a simplified version of adaptive cruise control algorithm in a simulated and real highway-inspired track. http://sedici.unlp.edu.ar:80/handle/10915/140244 Objeto de conferencia XXII Simposio Argentino de Inteligencia artificial (ASSAI 2021) - JAIIO 50 (Modalidad virtual) Actualmente, la experiencia de usuario es un aspecto fundamental en el diseño de sistemas informáticos, que debe considerar tanto la eficacia como la eficiencia en el uso y los resultados que devuelven dichos sistemas. Más aún, cuando se trata de búsquedas en la Web, esos factores son críticos ya que, debido al gran volumen de datos, distinguir la información útil no suele ser un hecho trivial. Por ende, la precisión y utilidad de la información que los usuarios demandan como resultado de una búsqueda son factores críticos. En base a estas cuestiones, este artículo describe un modelo de expansión de consultas basado en el modelado de tópicos. Específicamente, el modelo infiere relaciones entre palabras que conforman la consulta con el objetivo de encontrar nuevos términos que están implícitos o no han sido considerados por el usuario, y que luego pueden ser utilizados en consultas más profundas y contextualizadas. La implementación del modelo se explica en detalle y se ejemplifica su utilización con tres casos de prueba. Fri, 01 Jan 2021 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/140244 2021-01-01T00:00:00Z Actualmente, la experiencia de usuario es un aspecto fundamental en el diseño de sistemas informáticos, que debe considerar tanto la eficacia como la eficiencia en el uso y los resultados que devuelven dichos sistemas. Más aún, cuando se trata de búsquedas en la Web, esos factores son críticos ya que, debido al gran volumen de datos, distinguir la información útil no suele ser un hecho trivial. Por ende, la precisión y utilidad de la información que los usuarios demandan como resultado de una búsqueda son factores críticos. En base a estas cuestiones, este artículo describe un modelo de expansión de consultas basado en el modelado de tópicos. Específicamente, el modelo infiere relaciones entre palabras que conforman la consulta con el objetivo de encontrar nuevos términos que están implícitos o no han sido considerados por el usuario, y que luego pueden ser utilizados en consultas más profundas y contextualizadas. La implementación del modelo se explica en detalle y se ejemplifica su utilización con tres casos de prueba. http://sedici.unlp.edu.ar:80/handle/10915/140242 Objeto de conferencia XXII Simposio Argentino de Inteligencia artificial (ASSAI 2021) - JAIIO 50 (Modalidad virtual) La expansión de claves de búsqueda es una técnica que permite lograr mayor calidad en los resultados cuando se realizan búsquedas en Internet. Claramente, la elección de buenos términos es de suma importancia para obtener documentos acordes a las necesidades del usuario. Uno de los desafíos encontrados en el proceso de expansión de consultas, además de la elección de la fuente a consultar y cómo realizar esa consulta, es qué criterio utilizar para la selección de los términos que se consideren mejores candidatos para incluirlos en la expansión. En este artículo se describe un modelo que, a partir de una clave de búsqueda, detecta las entidades contenidas en ella, explora con qué otras entidades están relacionadas y construye grafos de conocimiento parciales basados en documentos de Wikipedia. Durante el proceso de creación de los grafos parciales se calcula la relevancia de cada nodo y se integran todos los nodos en un grafo final. Por último, los valores de relevancia obtenidos definen las mejores entidades que serán sugeridas como términos de expansión. Además de la descripción del modelo, se presentan dos ejemplos de utilización. Fri, 01 Jan 2021 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/140242 2021-01-01T00:00:00Z La expansión de claves de búsqueda es una técnica que permite lograr mayor calidad en los resultados cuando se realizan búsquedas en Internet. Claramente, la elección de buenos términos es de suma importancia para obtener documentos acordes a las necesidades del usuario. Uno de los desafíos encontrados en el proceso de expansión de consultas, además de la elección de la fuente a consultar y cómo realizar esa consulta, es qué criterio utilizar para la selección de los términos que se consideren mejores candidatos para incluirlos en la expansión. En este artículo se describe un modelo que, a partir de una clave de búsqueda, detecta las entidades contenidas en ella, explora con qué otras entidades están relacionadas y construye grafos de conocimiento parciales basados en documentos de Wikipedia. Durante el proceso de creación de los grafos parciales se calcula la relevancia de cada nodo y se integran todos los nodos en un grafo final. Por último, los valores de relevancia obtenidos definen las mejores entidades que serán sugeridas como términos de expansión. Además de la descripción del modelo, se presentan dos ejemplos de utilización. http://sedici.unlp.edu.ar:80/handle/10915/140241 Objeto de conferencia XXII Simposio Argentino de Inteligencia artificial (ASSAI 2021) - JAIIO 50 (Modalidad virtual) AI-Friendly is a web application that allows to easily perform binary classification with a neural network. It is free and allows network customization, making it a useful tool for machine learning courses. Fri, 01 Jan 2021 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/140241 2021-01-01T00:00:00Z AI-Friendly is a web application that allows to easily perform binary classification with a neural network. It is free and allows network customization, making it a useful tool for machine learning courses. http://sedici.unlp.edu.ar:80/handle/10915/140239 Objeto de conferencia XXII Simposio Argentino de Inteligencia artificial (ASSAI 2021) - JAIIO 50 (Modalidad virtual) La búsqueda de vías metabólicas tiene como objetivo encontrar secuencias de reacciones que permitan transformar un sustrato dado en un producto de interés. Esta tarea puede abordarse como un problema de búsqueda en grafos, usando la estructura molecular de los compuestos y una medida de similaridad entre las estructuras para guiar la búsqueda. Sin embargo, los enfoques basados en esta idea resultan inútiles cuando se carece de la estructura, lo que impide calcular la similaridad. Por su parte, las redes neuronales en grafos han demostrado ser de gran utilidad como extractores de características en datos con estructuras no-euclideanas. Aquí presentamos un modelo neuronal basado en grafos, capaz de aprender representaciones de los compuestos a partir de características simples y de la topología de la red que los conecta. Estas características son luego empleadas para inferir la similaridad, sin que sea necesaria la estructura de los mismos en el proceso. Los resultados muestran que el modelo infiere correctamente la similaridad entre compuestos con estructura conocida, y genera estimaciones razonables para compuestos con estructura desconocida. Fri, 01 Jan 2021 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/140239 2021-01-01T00:00:00Z La búsqueda de vías metabólicas tiene como objetivo encontrar secuencias de reacciones que permitan transformar un sustrato dado en un producto de interés. Esta tarea puede abordarse como un problema de búsqueda en grafos, usando la estructura molecular de los compuestos y una medida de similaridad entre las estructuras para guiar la búsqueda. Sin embargo, los enfoques basados en esta idea resultan inútiles cuando se carece de la estructura, lo que impide calcular la similaridad. Por su parte, las redes neuronales en grafos han demostrado ser de gran utilidad como extractores de características en datos con estructuras no-euclideanas. Aquí presentamos un modelo neuronal basado en grafos, capaz de aprender representaciones de los compuestos a partir de características simples y de la topología de la red que los conecta. Estas características son luego empleadas para inferir la similaridad, sin que sea necesaria la estructura de los mismos en el proceso. Los resultados muestran que el modelo infiere correctamente la similaridad entre compuestos con estructura conocida, y genera estimaciones razonables para compuestos con estructura desconocida. http://sedici.unlp.edu.ar:80/handle/10915/140237 Objeto de conferencia XXII Simposio Argentino de Inteligencia artificial (ASSAI 2021) - JAIIO 50 (Modalidad virtual). In this paper, we present a tool for possibilistic logic programming.This tool is a desktop-based, stand-alone application that assists a user in creating, editing and querying a possibly inconsistent possibilistic program. The tool computes all the arguments emerging from the program, the grounded extension based on Dung-style semantics, and it is capable of showing arguments and grounded extensions graphically. The language for programs is enriched with pragmas for allowing the user to con gure labels for necessity degrees, deciding if using transposes of strict rules, performing consistency checks within arguments, and appeal to the use of accrual of rules for building arguments. We describe its usage, architectural elements and we also provide experimental evaluation of its performance. Fri, 01 Oct 2021 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/140237 2021-10-01T00:00:00Z In this paper, we present a tool for possibilistic logic programming.This tool is a desktop-based, stand-alone application that assists a user in creating, editing and querying a possibly inconsistent possibilistic program. The tool computes all the arguments emerging from the program, the grounded extension based on Dung-style semantics, and it is capable of showing arguments and grounded extensions graphically. The language for programs is enriched with pragmas for allowing the user to con gure labels for necessity degrees, deciding if using transposes of strict rules, performing consistency checks within arguments, and appeal to the use of accrual of rules for building arguments. We describe its usage, architectural elements and we also provide experimental evaluation of its performance. http://sedici.unlp.edu.ar:80/handle/10915/140235 Objeto de conferencia XXII Simposio Argentino de Inteligencia artificial (ASSAI 2021) - JAIIO 50 (Modalidad virtual) Modern Natural Language Processing (NLP) models can achieve great results resolving di erent types of linguistic tasks. This is possible thanks to a high volume of internal parametersthat are optimized during the training phase. They allow to model high-level linguistic properties. For example, LSTM-based language models have the ability to nd long-term dependencies between words on a text, and use them to make predictions about upcoming words. Nevertheless, their complexity makes it hard to understand which features they use to generate predictions. The neurolinguistic eld faces a similar issue when studying how our brain processes language. For example, every adult reader has the ability to understand long texts and to make predictions of upcoming words. Nevertheless, our understanding on how these predictions are driven is limited. During the last decades, the study of eye movements during reading have shed some light on this topic, nding a relation between the time spent on a word (gaze duration) and its processing cost. Here, we aim to understand how LSTM-based models predict future words and these predictions relate with human predictions, tting statistical models commonly used in the neurolinguistic eld with gaze duration as the dependent variable. We found that an AWD-LSTM Language Model can partially model eye movements, with high overlap with both human-Predictability and lexical frequency. Interestingly, this last overlap is seen to depend on the training corpus, being lower when the model is ne-tuned with a corpus similar to the one used for testing. Fri, 01 Jan 2021 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/140235 2021-01-01T00:00:00Z Modern Natural Language Processing (NLP) models can achieve great results resolving di erent types of linguistic tasks. This is possible thanks to a high volume of internal parametersthat are optimized during the training phase. They allow to model high-level linguistic properties. For example, LSTM-based language models have the ability to nd long-term dependencies between words on a text, and use them to make predictions about upcoming words. Nevertheless, their complexity makes it hard to understand which features they use to generate predictions. The neurolinguistic eld faces a similar issue when studying how our brain processes language. For example, every adult reader has the ability to understand long texts and to make predictions of upcoming words. Nevertheless, our understanding on how these predictions are driven is limited. During the last decades, the study of eye movements during reading have shed some light on this topic, nding a relation between the time spent on a word (gaze duration) and its processing cost. Here, we aim to understand how LSTM-based models predict future words and these predictions relate with human predictions, tting statistical models commonly used in the neurolinguistic eld with gaze duration as the dependent variable. We found that an AWD-LSTM Language Model can partially model eye movements, with high overlap with both human-Predictability and lexical frequency. Interestingly, this last overlap is seen to depend on the training corpus, being lower when the model is ne-tuned with a corpus similar to the one used for testing. http://sedici.unlp.edu.ar:80/handle/10915/140171 Objeto de conferencia XXII Simposio Argentino de Inteligencia artificial (ASSAI 2021) - JAIIO 50 (Modalidad virtual) During dialogue, speakers have to rapidly represent different aspects of speech in order to respond and continue engaged. In particular, acoustic aspects indicate different intentions or emotions. Understanding how the brain represents those characteristics, both spatially and temporally, is important for developing better interactive systems. In this work, we analyse electroencephalography (EEG) data and audio recordings from ten participants engaged in a dialogue during a collaborative task. We aim to better understand how the envelope of the previous instants of the speech is encoded in the brain. Our results showed a good performance in predicting EEG signals, in particular in frontal electrodes. Moreover, our findings suggest that the brain encodes the speech envelope information with a latency between 0 and 100ms. These results are widely consistent between participants and also with previous work. The present work paves the way for studying brain representations of speech in natural scenarios with EEG. Fri, 01 Jan 2021 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/140171 2021-01-01T00:00:00Z During dialogue, speakers have to rapidly represent different aspects of speech in order to respond and continue engaged. In particular, acoustic aspects indicate different intentions or emotions. Understanding how the brain represents those characteristics, both spatially and temporally, is important for developing better interactive systems. In this work, we analyse electroencephalography (EEG) data and audio recordings from ten participants engaged in a dialogue during a collaborative task. We aim to better understand how the envelope of the previous instants of the speech is encoded in the brain. Our results showed a good performance in predicting EEG signals, in particular in frontal electrodes. Moreover, our findings suggest that the brain encodes the speech envelope information with a latency between 0 and 100ms. These results are widely consistent between participants and also with previous work. The present work paves the way for studying brain representations of speech in natural scenarios with EEG. http://sedici.unlp.edu.ar:80/handle/10915/140168 Objeto de conferencia XXII Simposio Argentino de Inteligencia artificial (ASSAI 2021) - JAIIO 50 (Modalidad virtual) Brain Computer Interfaces (BCIs) are useful devices that provide new ways of communication to people who have lost the capability of interacting with their environment. Although several paradigms have resulted in large improvements in the construction of BCIs, quite often they require great efforts from the patient or they are not able to generate natural and efficient interfaces. In that scenario, inner speech appears as a promising paradigm for tackling those problems. Nevertheless, the lack of publicly available databases largely precludes the analysis and development of methods for using this paradigm. In this work we use a recently released database to show that it is possible to classify and differentiate inner speech signals from signals acquired within other two well known paradigms. This is undoubtedly a first step in the search and construction of an inner speech based BCI. Fri, 01 Jan 2021 00:00:00 GMT http://sedici.unlp.edu.ar:80/handle/10915/140168 2021-01-01T00:00:00Z Brain Computer Interfaces (BCIs) are useful devices that provide new ways of communication to people who have lost the capability of interacting with their environment. Although several paradigms have resulted in large improvements in the construction of BCIs, quite often they require great efforts from the patient or they are not able to generate natural and efficient interfaces. In that scenario, inner speech appears as a promising paradigm for tackling those problems. Nevertheless, the lack of publicly available databases largely precludes the analysis and development of methods for using this paradigm. In this work we use a recently released database to show that it is possible to classify and differentiate inner speech signals from signals acquired within other two well known paradigms. This is undoubtedly a first step in the search and construction of an inner speech based BCI.