Search

La relación entre la Inteligencia Artificial y los videojuegos es muy estrecha. Por un lado, la IA puede usar los videojuegos como un campo de entrenamiento, y por otro, los videojuegos pueden implementar diversas técnicas de IA para generar comportamientos y contenido innovador. En este proyecto, se está trabajando en el desarrollo de un Agente Inteligente utilizando técnicas de aprendizaje por refuerzo, con el objetivo de que pueda desenvolverse en diferentes escenarios del videojuego Kart Gaming. Asimismo, el proyecto se enfoca en la creación de nuevos escenarios jugables para el mismo juego., The relationship between Artificial Intelligence and video games is very close. On the one hand, AI can use video games as a training ground, and on the other hand, video games can implement various AI techniques to generate innovative behaviour and content. In this project, we are working on the development of an Intelligent Agent using reinforcement learning techniques, so that it can perform in different scenarios of the video game Kart Gaming. The project also focuses on the creation of new playable scenarios for the same game.

Multi-Input Multi-Output (MIMO) radar systems have been extensively studied for their potential to enhance target tracking in complex environments. Recent advancements in the field of Artificial Intelligence (AI) and Machine Learning (ML) have significantly contributed to further improving the performance of these systems. This thesis focuses on studying a Reinforcement Learning (RL) model that employs an enhanced SARSA algorithm for MIMO tracking radar, starting from a RL approach of multi-target detection for cognitive radar (CR). A Monte Carlo method is used to generate randomness for noise, and targets are detected through a robust Wald type test statistic. We implemented a modified Double SARSA algorithm in the system model, as well as, added a new set of scenarios based on randomness in the position of the targets and their appearance in them. Finally, we evaluated the performance of both algorithms under this scenarios., Los sistemas de radar Multi-Input Multi-Output (MIMO) han sido ampliamente estudiados debido a su potencial para mejorar el seguimiento de objetivos en entornos complejos. Los recientes avances en el campo de la Inteligencia Artificial (IA) y el Aprendizaje Automático (AA) han contribuido significativamente a mejorar aún más el rendimiento de estos sistemas. Esta tesis se enfoca en estudiar un modelo de Aprendizaje por Refuerzo que utiliza un algoritmo SARSA mejorado para el seguimiento de radar MIMO, partiendo de un enfoque de RL para la detección de múltiples objetivos para radar cognitivo (CR). Se utiliza un método de Monte Carlo para generar aleatoriedad en el ruido, y los objetivos se detectan a través de una prueba robusta de estadística de tipo Wald. Además, se ha implementado un algoritmo Double SARSA modificado en el modelo radar, así como un nuevo conjunto de escenarios basados en la aleatoriedad en la posición de los objetivos y su aparición en ellos. Finalmente, se evalua el rendimiento de detección del radar para ambos algoritmos en estos escenarios., Els sistemes de radar Multi-Input Multi-Output (MIMO) han estat àmpliament estudiats a causa del seu potencial per millorar el seguiment d'objectius en entorns complexos. Els avenços recents en el camp de la Intel·ligència Artificial (IA) i l'Aprenentatge Automàtic (AA) han contribuït significativament a millorar encara més el rendiment d'aquests sistemes. Aquesta tesi s'enfoca en estudiar un model d'aprenentatge per reforç que utilitza un algorisme SARSA millorat pel seguiment de radar MIMO, el cual parteix d'un enfocament de RL per a la detecció de múltiples objectius per a radar cognitiu (CR), basats. S'utilitza un mètode de Monte Carlo per generar aleatorietat al soroll, i els objectius es detecten mitjançant una prova robusta d'estadística de tipus Wald. A més, s'ha implementat un algorisme Double SARSA modificat i un nou conjunt d'escenaris basats en l'aleatorietat en la posició dels objectius i la seva aparició. Finalment, s'avalua el rendiment de detecció del radar per a tots dos algorismes en aquests escenaris.

When integrating artificial agents into physical or digital environments that are shared with humans, agents are often equipped with opaque Machine Learning methods to enable adapting their behaviour to dynamic human needs and environment. This brings about agents that are also opaque and therefore hard to explain. In previous work, we show that we can reduce an opaque agent into an explainable Policy Graph (PG) which works accurately in multi-agent environments. Policy Graphs are based on a discretisation of the world into propositional logic to identify states, and the choice of which discretiser to apply is key to the performance of the reduced agent. In this work, we explore this further by 1) reducing a single agent into an explainable PG, and 2) enforcing collaboration between this agent and an agent trained from human behaviour. The human agent is computed by using GAIL from a series of human-played episodes, and kept unchanged. We show that an opaque agent created and trained to collaborate with the human agent can be reduced to an explainable, non-opaque PG, so long as predicates regarding collaboration are included in the state representation, by showing the difference in reward between the agent and its PG. Code is available at https://github.com/HPAI-BSC/explainable-agents-with-humans, This work has been partially supported by EU Horizon 2020 Project StairwAI (grant agreement No. 101017142)., Peer Reviewed, Postprint (published version)

Deep Learning (DL) techniques are at the heart of most future advanced software functions in Critical Autonomous AI-based Systems (CAIS), where they also represent a major competitive factor. Hence, the economic success of CAIS industries (e.g., automotive, space, railway) depends on their ability to design, implement, qualify, and certify DL-based software products under bounded effort/cost. However, there is a fundamental gap between Functional Safety (FUSA) requirements on CAIS and the nature of DL solutions. This gap stems from the development process of DL libraries and affects high-level safety concepts such as (1) explainability and traceability, (2) suitability for varying safety requirements, (3) FUSA-compliant implementations, and (4) real-time constraints. As a matter of fact, the data-dependent and stochastic nature of DL algorithms clashes with current FUSA practice, which instead builds on deterministic, verifiable, and pass/fail test-based software. The SAFEXPLAIN project tackles these challenges and targets by providing a flexible approach to allow the certification - hence adoption - of DL-based solutions in CAIS building on: (1) DL solutions that provide end-to-end traceability, with specific approaches to explain whether predictions can be trusted and strategies to reach (and prove) correct operation, in accordance to certification standards; (2) alternative and increasingly sophisticated design safety patterns for DL with varying criticality and fault tolerance requirements; (3) DL library implementations that adhere to safety requirements; and (4) computing platform configurations, to regain determinism, and probabilistic timing analyses, to handle the remaining non-determinism., The research leading to these results has received funding from the Horizon Europe Programme under the SAFEXPLAIN Project (www.safexplain.eu), grant agreement num. 101069595. BSC authors have also been supported by the Spanish Ministry of Science and Innovation under grant PID2019- 107255GBC21/AEI/10.13039/501100011033., Peer Reviewed, "Article signat per 22 autors/es: Jaume Abella, Jon Perez, Cristofer Englund, Bahram Zonooz, Gabriele Giordana, Carlo Donzella, Francisco J. Cazorla, Enrico Mezzetti, Isabel Serra, Axel Brando, Irune Agirre, Fernando Eizaguirre, Thanh Hai Bui, Elahe Arani, Fahad Sarfraz, Ajay Balasubramaniam, Ahmed BadarIlaria Bloise, Lorenzo Feruglio, Ilaria Cinelli, Davide Brighenti, Davide Cunial", Postprint (author's final draft)

This paper presents for the first time successful results of active flow control with multiple independently controlled zero-net-mass-flux synthetic jets. The jets are placed on a three-dimensional cylinder along its span with the aim of reducing the drag coefficient. The method is based on a deep-reinforcement-learning framework that couples a computational-fluid-dynamics solver with an agent using the proximal-policy-optimization algorithm. We implement a multi-agent reinforcement- learning framework which offers numerous advantages: it exploits local invariants, makes the control adaptable to different geometries, facilitates transfer learning and cross-application of agents and results in significant training speedup. In this contribution we report significant drag reduction after applying the DRL-based control in three different configurations of the problem., Ricardo Vinuesa acknowledges funding by the ERC through Grant No. “2021-CoG-101043998, DEEPCONTROL”., Peer Reviewed, Postprint (published version)

Agent-Based Simulation is a suitable approach used now-a-days to simulate and analyze complex societal environments and scenarios. Current Agent-Based Simulation frameworks either scale quite well in computation but implement very simple reasoning mechanisms, or employ complex reasoning systems at the expense of scalability. In this paper we present our work to extend an agent-based HPC platform, enabling goal-driven agents with HTN planning capabilities to scale and run parallelly. Our extension includes preferences over their objectives, preferences over their plans, actions, and moral values. We show the expresiveness of the extended platform with a sample scenario., This work has been partially supported by EU Horizon 2020 Project StairwAI (grant agreement No.101017142)., Peer Reviewed, Postprint (published version)

We propose a hierarchical framework for collaborative intelligent systems. This framework organizes research challenges based on the nature of the collaborative activity and the information that must be shared, with each level building on capabilities provided by lower levels. We review research paradigms at each level, with a description of classical engineering-based approaches and modern alternatives based on machine learning, illustrated with a running example using a hypothetical personal service robot. We discuss cross-cutting issues that occur at all levels, focusing on the problem of communicating and sharing comprehension, the role of explanation and the social nature of collaboration. We conclude with a summary of research challenges and a discussion of the potential for economic and societal impact provided by technologies that enhance human abilities and empower people and society through collaboration with intelligent systems., This work was supported in part by the MIAI Multidisciplinary AI Institute at the Universite Grenoble Alpes (MIAI@Grenoble Alpes - ANR-19-P3IA-0003), in part by the EU H2020 ICT AI4EU under Grant 825619, and in part by the EU H2020 project Humane AI Net under Grant 952026., Peer Reviewed, Postprint (author's final draft)

Se propone el desarrollo de sistemas y algoritmos de aprendizaje reforzado profundo para entornos de vehículos autónomos. Para ello se propone inicialmente realizar una búsqueda bibliográfica sobre el uso de esta técnica de aprendizaje reforzado profundo para aplicaciones futuras de vehículos autónomos. Otro elemento básico de este proyecto será el desarrollo de herramientas de aprendizaje reforzado profundo, para mejorar en lo posible, la capacidad de aprendizaje del vehículo, la capacidad de adaptación a un entorno cambiante, y su capacidad final de decidir y realizar u, The project explores the possibilities offered by reinforcement learning in the field of robotics with the vision of guiding robots in changing environments with collision avoidance through potential fields. For this, the DDPG, TD3, SAC and PPO reinforcement learning algorithms are implemented through the Matlab Toolbox "Reinforcement Learning" with the aim of carrying out a comparative study on which of them is the most optimal for different configurations of environments and parameters, with the help of training graphs and statistical tables. Also, potential fields have been developed in this project, demonstrating to be a suitable tool for guiding robots in changing environments, and even to implement multi agent scenarios, avoiding collisions among them and enhancing collaboration., El projecte explora les possibilitats que ofereix l'aprenentatge per reforç en l'àmbit de la robòtica amb la visió de guiar robots a través d'entorns canviants amb evitació de col·lisions mitjançant camps de potencials. Per això s'implementen els algorismes d'aprenentatge per reforç DDPG, TD3, SAC i PPO per mitjà de la Toolbox de Matlab Reinforcement Learning amb l'objectiu de fer un estudi comparatiu sobre quin d'ells és el més òptim per a diferents configuracions d'entorns i paràmetres; tot això amb l'ajuda de gràfiques d'entrenament i taules estadístiques. Així mateix, s'han desenvolupat camps potencials en aquest projecte, demostrant ser una eina adequada per a guiar robots en entorns canviants, i fins i tot per implementar escenaris multiagent, evitant col·lisions entre ells i potenciant la col·laboració., El proyecto explora las posibilidades que ofrece el aprendizaje por refuerzo en el ámbito de la robótica con la visión de guiar a robots a través de entornos cambiantes con evitación de colisiones mediante campos potenciales. Para ello se implementan los algoritmos de aprendizaje por refuerzo DDPG, TD3, SAC y PPO por intermedio de la Toolbox de Matlab "Reinforcement Learning" con el objetivo de realizar un estudio comparativo sobre cuál de ellos es el más óptimo para diferentes configuraciones de entornos y parámetros; todo ello con la ayuda de gráficas de entrenamiento y tablas estadísticas. Además, en este proyecto se han desarrollado campos potenciales, demostrando ser una herramienta adecuada para guiar robots en entornos cambiantes, e incluso implementar escenarios multiagente, evitando colisiones entre ellos y potenciando la colaboración.

Como estudiante, en muchas ocasiones he tenido la necesidad de preguntar sobre cuestiones académicas que no lograba encontrar. Y el hecho de que las entidades que pueden responderlas estén saturadas en ocasiones genera una sensación de frustración en el alumno que termina en un mayor descontento y en más saturación de las entidades en cuestión. Es por ello que el objetivo de este proyecto es crear un bot fácilmente escalable que permita a los alumnos facilitar sus problemas del día a día. Si bien no sustituye a dichas entidades, puede reducir su carga de trabajo mediante respuestas más automáticas a cuestiones que resulten más 'triviales' para el alumno. Juntando todo lo que el alumno pueda interesarle en un único menú y con la posibilidad de contactar desde allí con las entidades que necesite si se diera el caso., As a student, on many occasions, I have needed to ask academic questions that I could not find. And the fact that the entities that can answer them are sometimes saturated generates a feeling of frustration in the student that ends in greater discontent and more saturation of the entities in question. That is why the objective of this project is to create an easily scalable bot that allows students to facilitate their day-to-day problems. Although it does not replace these entities, it can reduce their workload through more automatic answers to questions that are more 'trivial' for the student. Bringing together everything that the student may be interested in in a single menu and with the possibility of contacting from there with the entities that they need if the case arises.

La Inteligencia Artificial es un área que trata de evolucionar constantemente con el fin de poder abordar cualquier problema, consiguiendo que hoy en día prácticamente todo pueda estar automatizado. En nuestro trabajo vamos a centrarnos en el aprendizaje por refuerzo, un área que cobra cada vez más fuerza dentro de este ámbito. Hace relativamente poco, Unity presento un kit de herramientas para el aprendizaje automático llamado ml-agents. Esta herramienta nos permite llevar a cabo proyectos muy sencillos así como otros de mucha complejidad, y todo esto sin la necesidad de tener una gran experiencia en el ámbito. En este proyecto vamos a diseñar e implementar un sistema complejo donde un agente aprenda a navegar por entornos naturales de la forma mas realista posible. Buscaremos explotar al máximo las capacidades de esta herramientas, tratando que las decisiones que toma nuestro agente sean lógicas y coherentes en función del escenario que tenga delante., Artificial Intelligence is an area that is constantly evolving in order to be able to tackle any problem, so that nowadays practically everything can be automated. In our work we are going to focus on reinforcement learning, an area that is gaining more and more strength in this field. Relatively recently, Unity introduced a machine learning toolkit called ml-agents. This tool allows us to carry out very simple projects as well as others of great complexity, and all this without the need of having a great deal of experience in the field. In this project we are going to design and implement a complex system where an agent learns to navigate through natural environments in the most realistic way possible. We will seek to exploit the capabilities of these tools to the maximum, trying to ensure that the decisions made by our agent are logical and coherent depending on the scenario he is faced with.

The rise of social networks and the increase of notifications we receive on our phones has had a negative impact on the way people work. Distractions have become a more serious problem on such times when productivity is the center of attention of many project managers. This project has analyzed this issue, and proposed a modern and flexible solution based on a conversational agent. It has implemented and evaluated a mobile application designed to support focused work, obtaining results that suggest the validation and effectiveness of a work assistant to increase productivity and well-being.

Supreme és una marca d'skate que comercialitza roba i altres tipus de productes. En els últims anys s'ha tornat molt popular, sobretot entre els adolescents, i ha esdevingut tot un fenomen de masses i un actor influent a la indústria de la moda. La característica més distintiva de Supreme és el seu model de negoci. Els productes de la marca no estan disponibles per comprar en qualsevol moment, en lloc d'això, la marca divideix les seves col·leccions en diversos llançaments («drops») setmanals. Els articles es llancen els dijous de cada setmana, a una hora determinada, amb un estoc molt limitat i quan s'esgoten no tornen a estar disponibles mai més. La majoria d'articles s'esgoten en qüestió de segons, això vol dir que hi ha una diferència notable entre l'oferta de productes i la seva demanda. Fet que ha propiciat l'aparició d'un gran mercat de revenda. Inicialment, els clients utilitzaven scripts rudimentaris i extensions de navegador per aconseguir un avantatge respecte els altres compradors i ser els primers en passar per caixa. Poc a poc, aquests mètodes van anar evolucionant fins als complexos bots de compra que existeixen avui en dia. Aquests bots són capaços de realitzar tot el procés de compra de forma automàtica i d'obtenir el producte desitjat per l'usuari sense la necessitat d'interacció humana. L'ús de bots de compra a Supreme està tan estès que si un usuari vol adquirir un dels articles més cobdiciats d'un llançament, llavors és imprescindible que n'utilitzi un. Al llarg d'aquest treball es descriu el desenvolupament d'un bot de compra basat en emulació del navegador, començant per l'estudi del context del projecte, passant per la definició de l'abast i la metodologia utilitzada, seguint amb l'especificació i el disseny de l'aplicació i acabant amb l'explicació de la implementació i la valoració dels resultats. Els bots basats en emulació interactuen amb la pàgina de Supreme a través del navegador, utilitzant els elements visuals de la interfície web, co, Supreme is a skate brand that sells clothing and other types of products. In recent years it has become very popular, especially among teenagers, and has become a mass phenomenon and an influential actor in the fashion industry. Supreme's most distinctive feature is its business model. The products of the brand are not available all the time, instead the brand divides its collections into several weekly releases ("drops"). The items are released on Thursdays of each week, at a certain time, with a very limited stock and when they run out they are never available again. Most items sell out in a matter of seconds, which means that there is a noticeable difference between product supply and demand. This fact has led to the appearance of a large resale market. Initially, customers used rudimentary scripts and browser extensions to gain an advantage over other shoppers and be the first to checkout. Gradually, these methods evolved into the complex shopping bots that exist today. These bots are capable of carrying out the entire purchase process automatically and obtain the product desired by the user without human interaction. The use of shopping bots on Supreme is so widespread that if a user wants to purchase one of the most coveted items from a drop, then using one is a must. This final thesis describes the development of a shopping bot that is based on browser emulation, starting with the study of the project context, going through the definition of the project scope and the methodology used, continuing with the specification and design of the application and ending with the explanation of the implementation and the evaluation of the results. Emulation-based bots interact with the Supreme page through the browser, using the visual elements of the web interface, as if they were normal buyers. The advantage of this type of bots is that they have the fingerprint of a real browser and they pass unnoticed by the control systems. To implement the application, Java has been

Recent advancements in crowd simulation unravel a wide range of functionalities for virtual agents, delivering highly-realistic,natural virtual crowds. Such systems are of particular importance to a variety of applications in fields such as: entertainment(e.g., movies, computer games); architectural and urban planning; and simulations for sports and training. However, providingtheir capabilities to untrained users necessitates the development of authoring frameworks. Authoring virtual crowds is acomplex and multi-level task, varying from assuming control and assisting users to realise their creative intents, to deliveringintuitive and easy to use interfaces, facilitating such control. In this paper, we present a categorisation of the authorable crowdsimulation components, ranging from high-level behaviours and path-planning to local movements, as well as animation andvisualisation. We provide a review of the most relevant methods in each area, emphasising the amount and nature of influencethat the users have over the final result. Moreover, we discuss the currently available authoring tools (e.g., graphical userinterfaces, drag-and-drop), identifying the trends of early and recent work. Finally, we suggest promising directions for futureresearch that mainly stem from the rise of learning-based methods, and the need for a unified authoring framework., This work has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska Curie grant agreement No 860768 (CLIPE project). This project has received funding from the European Union’s Horizon 2020 Research and Innovation Programme under Grant Agreement No 739578 and the Government of the Republic of Cyprus through the Deputy Ministry of Research, Innovation and Digital Policy, Peer Reviewed, Postprint (author's final draft)

Donat l'enorme creixement de les tecnologies blockchain a finals de 2021. En particular un dels seus actius anomenat NFT, un actiu que és una nova forma d'escassetat digital. Aquest projecte pretén estudiar el seu mercat encara altament volàtil i desenvolupar tant un DataFrame com un bot, capaços de predir el rendiment economic d'una col·lecció de NFTs en el futur proper. El projecte es centra en el mercat principal de NFT Opensea i concretament en les col·leccions de l'ecosistema Ethereum, que té el 80% del mercat a principis de 2022. Degut a la seva novetat, fins ara s'han fet poques aplicacions i estudis sobre aquest nou mercat. Però això no ha impedit que aquest nou actiu augmenti ràpidament en popularitat des de els 10.000 usuaris fins a més d'1,5 milions només en l'últim any. És cert que a mesura que creix l'interès es publicaran més informació i projectes, però cap semblant a aquesta tesi a la data d'inici. Quan s'intenta fer un anàlisi de dades la mida de la mostra és un factor clau, però aquest és un escàs recurs quan es parla de transaccions NFT. Per això per fer tota l'anàlisi i proves d'aquest projecte, s'han seleccionat 36 de les col·leccions més venudes del mercat. Per crear aquest conjunt només s'han triat col·leccions principals, descartant col·leccions secundàries de marques ja consolidades en l'espai. A més, per garantir dades homogènies, tots els DataFrames emmagatzemen informació dels mateixos períodes des de l'1 de desembre de 2021 fins al 31 de març de 2022. Mitjançant l'API d'Opensea, la majoria de les dades necessàries s'han extret directament de la blockchain. Milers de transaccions de cada col·lecció NFT s'han processat i emmagatzemat en dataFrames, produint fitxers .csv facilment llegibles emmagatzemats en un directori per a un ús futur. Finalment, pel que fa al bot, s'ha utilitzat l'algoritme d'aprenentatge automàtic per a la predicció de dades LSTM (Long Short Term Memory). Aquest s'utilitza habitualment per a la predicció de la borsa i, Given the enormous growth of blockchain technologies at the end of 2021. Particularly one of its assets called NFTs, an asset that is a new form of digital scarcity. This project aims to study their still highly volatile market and develop both a dataFrame and a bot, capable of predicting a NFT collection performance in the near future. Project focuses on the NFT main marketplace Opensea and specifically the collections in the Ethereum ecosystem, which has 80% of the market as of early 2022. Because of its novelty few applications and studies have been done on this field to date. But this hasn't stopped this new asset to rapidly rise in popularity from a small 10k users to more than 1.5M only in the last year. It is true that as interest grows more and more information and projects are being published, but none similar to this thesis as of start date. When trying to do a data analysis sample size is a key factor, but a scarce resource when talking about NFT transactions. Because of this to make all the analysis and testing of this project 36 from the best-selling collections in the market have been selected. To create this set only main collections have been chosen, discarding secondary collections from already established brands. Moreover, to ensure homogeneous data, all dataFrames store information from the same periods from December 1st, 2021 to March 31, 2022. Using the Opensea API, most of the data needed has been taken directly from the blockchain. Thousands of transactions from each NFT collection have then been processed and stored into dataFrames, producing readable .csv files stored into a directory for future use. Lastly regarding the bot, machine learning algorithm for data prediction LSTM (Long Short Term Memory) have been used. Commonly used for stock market prediction, this algorithm has been selected as the most promising one for this project. Models have been trained and tested using subsets of each collection dataFrames. After that, further examina

Path finding for autonomous agents has been traditionally driven by finding optimal paths, typically by using A* search or any of its variants. When it comes to simulating virtual humanoids, traditional approaches rarely consider aspects of human memory or orientation. In this work, we propose a new path finding algorithm, inspired by current research regarding how the brain learns and builds cognitive maps. Our method represents the space as a hexagonal grid with counters, based on brain research that has investigated how memory cells are fired. Our path finder then combines a method for exploring unknown environments while building such a cognitive map, with an A* search using a modified heuristic that takes into account the cognitive map. The resulting paths show how as the agent learns the environment, the paths become shorter and more consistent with the optimal A* search. Moreover, we run a perceptual study to demonstrate that the viewers could successfully identify the intended level of knowledge of the simulated agents. This line of research could enhance the believability of autonomous agents’ path finding in video games and other VR applications., This work was partly funded by the Spanish Ministry of Economy, Industry and Competitiveness under Grant No. TIN2017- 88515-C2-1-R., Peer Reviewed, Postprint (author's final draft)

High-level thinking, such as decision-making, problem solving, and critical thinking, is an essential skill that people need to learn and apply in their daily lives. Such thinking is often complex, and people can traditionally get help from others for task completion. For example, students can get guidance from instructors on how to read papers critically. However, such human experts are not always available. Intelligent agents in a robot or bot form can mitigate this issue by socially offering real-time assistance to users. Yet, it is challenging to design the agents that can appropriately support users in high-level thinking tasks, because interacting with the agents may distract users from the main tasks which could require a lot of mental effort. Researchers of these intelligent agents need to answer: where can the agents offer help; how to design their behaviors; how to develop the agents for design validation; and how to evaluate the proposed design. My thesis research investigates the design and user experience of intelligent agents and their manner (i.e., way of behaving) for supporting users’ high-level thinking tasks. To address the design questions mentioned above, I adopt a set of human-centered approaches that first discover the usage scenarios of the agents by emphasizing with targeted users. Then I design the agents and their manners by summarizing human assistants’ behavioral patterns and existing technological practices. Next, I develop the agent prototypes and evaluate our design via user studies with targeted users. I adopt this approach to design and evaluate intelligent agents in three high-level thinking support contexts separately, i.e., decision-making, problem solving, and critical thinking. In the first study, we learn from human workers’ behaviors and robot’s autonomy to model the service robot’s proactivity (low, medium, high) in decision-making support tasks. Our experiment in a simulated shopping scenario shows that a highly proactive r

En aquest projecte es desenvolupa un plugin que integra Botonic (framework per desenvolupar aplicacions conversacionals) amb un CMS anomenat Directus. El problema que es vol resoldre amb aquesta integració és la gestió dels continguts que es mostren a les aplicacions conversacionals. D'aquesta manera, enlloc de guardar els continguts en fitxers de la propia aplicació, aquesta informació es guarda en bases de dades i s'obté des del CMS mitjançant una API. Les principals millores que s'hi presenten amb aquesta solució són: no és necessari deployar l'aplicació de nou cada vegada que es vulgui fer un canvi en els continguts, usuaris externs, sense haver de tocar codi, poden manipular els continguts i els canvis s'hi veuen reflectits instantàniament, els continguts es poden organitzar i gestionar d'una manera més endreçada, hi ha molta més flexibilitat i possibilitats per crear continguts de diferents formes, amb diferents atributs i diferents funcionalitats, fan que el codi de l'aplicació quedi molt més nét i llegible. Per a poder triar la millor solució, s'ha realitzat un estudi i anàlisi exhaustius del mercat. Un cop s'ha escollit quin CMS encaixa el màxim possible amb les necessitats que s'hi presenten, s'ha desenvolupat el plugin d'una manera estructurada i seguint el principi de programació SOLID. Per cada funcionalitat s'han anat creant tests unitaris per assegurar que el codi funcioni correctament i per disminuir la possibilitat de l'existència d'algun bug o error. Els resultats han estat satisfactoris i s'ha pogut resoldre el problema plantejat. S'ha aconseguit integrar el CMS, podent guardar-hi la informació i obtenint-la per mostrar-la en una aplicació conversacional., In this project a plugin has been developed in order to integrate Botonic (framework for developing conversational applications) with a CMS called Directus. The problem to be solved with this integration is the management of the content displayed in conversational applications. In this way, instead of saving the contents in files of the application itself, this information is saved in databases and is obtained from the CMS through an API. The main improvements that are presented with this solution are: it is not necessary to deploy the app every time you want to make a change to the content, external users, without having to touch code, can manipulate the contents and the changes are reflected instantly, the contents can be organized and managed in a more orderly way, there is much more flexibility and possibilities to create content in different ways, with different attributes and different functionalities, make the application code much cleaner and more readable. In order to choose the best solution, an exhaustive study and analysis of the market has been carried out. Once it has been chosen which CMS fits the maximum possible with the needs that are presented, the plugin has been developed in a structured way and following the SOLID programming principle. Unit tests have been created for each functionality to ensure that the code works correctly and to reduce the possibility of the existence of a bug or error. The results have been satisfactory and the problem raised has been solved. The CMS has been integrated, being able to save the information and obtain it to display it in a conversational application.

As long as critical levels of vaccination have not been reached to ensure heard immunity, and new SARS-CoV-2 strains are developing, the only realistic way to reduce the infection speed in a population is to track the infected individuals before they pass on the virus. Testing the population via sampling has shown good results in slowing the epidemic spread. Sampling can be implemented at different times during the epidemic and may be done either per individual or for combined groups of people at a time. The work we present here makes two main contributions. We first extend and refine our scalable agent-based COVID-19 simulator to incorporate an improved socio-demographic model which considers professions, as well as a more realistic population mixing model based on contact matrices per country. These extensions are necessary to develop and test various sampling strategies in a scenario including the 62 largest cities in Spain; this is our second contribution. As part of the evaluation, we also analyze the impact of different parameters, such as testing frequency, quarantine time, percentage of quarantine breakers, or group testing, on sampling efficacy. Our results show that the most effective strategies are pooling, rapid antigen test campaigns, and requiring negative testing for access to public areas. The effectiveness of all these strategies can be greatly increased by reducing the number of contacts for infected individual., This work has been supported by the Carlos III Institute of Health under the project grant 2020/00183/001, the project grant BCV-2021-1-0011, of the Spanish Supercomputing Network (RES) and the European Union’s Horizon 2020 JTI-EuroHPC research and innovation program under grant agreement No 956748. The role of all study sponsors was limited to financial support and did not imply participation of any kind in the study and collection, analysis, and interpretation of data, nor in the writing of the manuscript., Peer Reviewed, Postprint (published version)

There may be a lot of abusive behaviour in conversations between teenagers, which take place through social media. In this project, we develop classifiers to find out which texts present abuse such as violence, sexual behaviour, disorder, anxiety... For this reason, we use different classifiers that take advantage of the most recent techniques based on language modeling using deep neural networks, namely Transformers. But we find it difficult to obtain supervised data, i.e. data tagged with the feeling it contains. Luckily, thanks to the countless sources of information available on the Internet, we can get unsupervised data and examples of text on the Internet. Consequently, this project aims to explore the possibility of using unsupervised data, beyond using a supervised set to train a system in the task of sentiment classification. More specifically in short texts, in informal contexts and in Catalan and Spanish. We have a set of about 200,000 supervised samples and a set of 300,000 unsupervised samples. We will introduce and study different models of Transformers (BERT, GAN- BERT, XLM-RoBERTa and ELECTRA) and their performance in the work of sentiment analysis. To finish with, based on models that achieve a better score in the binary classification task of feelings, i.e. with ELECTRA and XLM-RoBERTa, we construct a multi- label classifier., En las conversaciones entre adolescentes, que se dan por medio de las redes sociales, pueden existir muchos comportamientos abusivos. En este proyecto desarrollamos clasificadores que detecten qué textos presentan algún abuso como violencia, conductas sexuales, desórdenes, ansiedad... Por este motivo, usamos diferentes clasificadores que aprovechan las técnicas más recientes basadas en modelado de lenguaje mediante redes neuronales profundas, concretamente Transformers. Pero nos encontramos con la dificultad de obtener datos supervisados, es decir datos etiquetados con el sentimiento que contiene. Por suerte, gracias a la infinidad de fuentes de información de las que disponemos en la red, podemos conseguir datos y ejemplos de texto no supervisado a internet. Por consiguiente, este proyecto quiere explorar la posibilidad de utilizar datos no supervisados, más allá de usar un conjunto de datos supervisado para entrenar un sistema en la tarea de clasificación de sentimientos. Más concretamente en textos cortos, en contextos informales y de lengua catalana y castellana. Contamos con un set de cerca de 200.000 muestras supervisadas y un set de 300.000 muestras no supervisadas. Introduciremos y estudiaremos diferentes modelos de Transformers (BERT, GAN-BERT, XLM-RoBERTa y ELECTRA) y su rendimiento en la tarea de análisis de sentimientos. Finalmente, a partir de los modelos que consiguen una mejor puntuación en la tarea de clasificación binaria de sentimientos, es decir con ELECTRA y XLM-RoBERTa, construimos un clasificador multi-etiqueta., En les converses entre adolescents, que es donen per mitjà de les xarxes socials, poden existir molts comportaments abusius. En aquesta tesi desenvolupem classificadors que detectin quins textos presenten algun abús com violència, conductes sexuals, desordres, ansietat... Per aquest motiu, faig servir diferents classificadors que aprofiten les tècniques més recents basades en modelat de llenguatge mitjançant xarxes neuronals profundes, concretament Transformers. Però ens trobem amb la dificultat d?obtenir dades supervisades, és a dir dades etiquetades amb el sentiment que conté. Per sort, gràcies a la infinitat de fonts d?informació de les que disposem a la xarxa, podem aconseguir dades i exemples de text no supervisat a internet. Per consegüent, aquest projecte vol explorar la possibilitat d'utilitzar dades no supervisades, més enllà de fer servir un set supervisat per entrenar un sistema en la tasca de classificació de sentiments. Més concretament en textos curts, en contextos informals i de llengua catalana i castellana. Comptem amb un set de prop de 200.000 mostres supervisades i un set de 300.000 mostres no supervisades. Introduirem i estudiarem diferents models de Transformers (BERT,GAN-BERT,XLM-Roberta i Electra) i el seu rendiment en la tasca d'anàlisi de sentiments. Finalment, a partir dels models que aconsegueixen una millor puntuació en la tasca de classificació binària de sentiments, és a dir amb Electra i XLM-Roberta, construïm un classificador multi-label

Although smart devices markets are increasing their sales figures, their computing capabilities are not sufficient to provide good-enough-quality services. This paper proposes a solution to organize the devices within the Cloud-Edge Continuum in such a way that each one, as an autonomous individual –Agent–, processes events/data on its embedded compute resources while offering its computing capacity to the rest of the infrastructure in a Function-as-a-Service manner. Unlike other FaaS solutions, the described approach proposes to transparently convert the logic of such functions into task-based workflows backing on task-based programming models; thus, agents hosting the execution of the method generate the corresponding workflow and offloading part of the workload onto other agents to improve the overall service performance. On our prototype, the function-to-workflow transformation is performed by COMPSs; thus, developers can efficiently code applications of any of the three envisaged computing scenarios – sense-process-actuate, streaming and batch processing – throughout the whole Cloud-Edge Continuum without struggling with different frameworks specifically designed for each of them., This work has been supported by the Spanish Government (PID2019-107255GB), by Generalitat de Catalunya (contract 2014-SGR-1051), and by the European Commission through the Horizon 2020 Research and Innovation program under Grant Agreement No. 101016577 (AI-SPRINT project)., Peer Reviewed, Postprint (author's final draft)

The advancement on explainability techniques is quite relevant in the field of Reinforcement Learning (RL) and its applications can be beneficial for the development of intelligent agents that are understandable by humans and are able cooperate with them. When dealing with Deep RL some approaches already exist in the literature, but a common problem is that it can be tricky to define whether the explanations generated for an agent really reflect the behaviour of the trained agent. In this work we will apply an approach for explainability based on the creation of a Policy Graph (PG) that represents the agent’s behaviour. Our main contribution is a way to measure the similarity between the explanations and the agent’s behaviour, by building another agent that follows a policy based on the explainability method and comparing the behaviour of both agents., Peer Reviewed, Postprint (published version)

Mobile applications have become a commodity in multiple daily scenarios. Their increasing complexity has led mobile software ecosystems to become heterogeneous in terms of hardware specifications, features and context of use, among others. For their users, fully exploiting their potential has become challenging. While enacting software systems with adaptation mechanisms has proven to ease this burden from users, mobile devices present specific challenges related to privacy and security concerns. Nevertheless, rather than being a limitation, users can play a proactive role in the adaptation loop by providing valuable feedback for runtime adaptation. To this end, we propose the use of chatbots to interact with users through a human-like smart conversational process. We depict a work-in-progress proposal of an end-to-end framework to integrate semi-automatic adaptation mechanisms for mobile applications. These mechanisms include the integration of both implicit and explicit user feedback for autonomous user categorization and execution of enactment action plans. We illustrate the applicability of such techniques through a set of scenarios from the Mozilla mobile applications suite. We envisage that our proposal will improve user experience by bridging the gap between users’ needs and the capabilities of their mobile devices through an intuitive and minimally invasive conversational mechanism., This work has been partially supported by AGAUR, code 2017-SGR-1694. The corresponding author gratefully acknowledges the Universitat Politècnica de Catalunya and Banco Santander for the financial support of his predoctoral grant FPI-UPC., Peer Reviewed, Postprint (author's final draft)