loading subjects...
Ebook: Artificial Intelligence Research and Development
loading...
Artificial Intelligence (AI) is one of the hottest areas of research with hundreds of potentially real life applications, and one of the disciplines that is most intertwined with all sorts of fields related to human experience. A wide range of disciplines, sectors and technologies could benefit from the results of AI breakthroughs; from robotics to search engines; healthcare or environment; screening, monitoring or supervision techniques. The present book compiles the 33 accepted papers presented at the 13th International Conference of the Catalan Association for Artificial Intelligence. Although most of the authors come from Catalan-speaking regions, the Conference has a long-standing collaboration with the AI community and this is reflected in the variety and interest of this work. How do Autonomy and AI relate to Aerospace Systems? Did you know that less is also more when it comes to Cultural Algorithms and that AI could also help with the treatment of major depressions? Or that AI could improve your city transport system? This book answers all questions and is especially addressed to IT engineers and software experts and, more generally, to everyone interested in AI technologies and what they have to offer for a better brighter future.
This volume contains the original contributions accepted for presentation at the 13th International Conference of the Catalan Association for Artificial Intelligence (CCIA'2010), held in the cave surrounded town of L'Espluga de Francolí, on October 20–22, 2010.
The Catalan Association for Artificial Intelligence (ACIA)
AClA is a member of ECCAI, the European Coordinating Committee for Artificial Intelligence (http://www.acia.org). Its name is an acronym of Associació Catalana d'Intel·ligència Artificial.
The collaboration between ACIA members and the world AI community at large is well-established for many decades now. The papers submitted to the conference reflect that long-standing collaboration; most of its authors come from Catalan-speaking regions of Spain, but also from neighboring countries such as France and Italy, other European Union countries such as the The Netherlands and United Kingdom, and farther countries such as Cuba, Colombia, Brazil and Korea. Over the years, CCIA has become steadily more international as the conference settled itself.
Each of the 43 papers submitted to CCIA'2010 was carefully reviewed by at least 3 program committee members and finally 35 papers have been accepted for presentation during the conference. Of those contributions, the 33 original ones are presented in this volume as well as a brief summary of the two invited talks. The accepted papers, from most of the areas of AI, were divided into 25 oral presentations and 10 posters for organizational purposes, but no difference has been made in the proceedings.
Of all the areas of Computer Science, AI is the one most intertwined with all sorts of disciplines dealt with in the human experience, often employing lessons from one discipline to achieve a task in another one. This diversity is also reflected in the invited talks in this conference: one involving web search, Towards Language-Competent Web Search, presented by Veronica Dahl, Professor at Simon Fraser University, Canada, and the other involving aerospace systems, Autonomy and Artificial Intelligence in Aerospace Systems, presented by Juan de Dalmau, Director of the Aerospace Research and Technology Centre (CTAE), Catalonia.
We would like to express our sincere gratitude to all the authors, who chose this conference to send their work, and to the members of the organizing committee, who took care of the uncountable detailed tasks necessary to have a successful conference. We heartily thank all the members of the Scientific Committee and other reviewers for the hard task of judging the submitted papers. Special thanks go also to the plenary speakers for their effort in preparing very interesting lectures, to Sandra Sandri, former PC chair in CCIA'2009, for her help during this year and, last but not least, to the president of the ACIA, Núria Agell, for her kind and constant support.
L'Espluga de Francolí, October 2010,
René Alquézar, Universitat Politècnica de Catalunya,
Antonio Moreno, Universitat Rovira i Virgili,
Josep Aguilar, Laboratoire d'Architecture et d'Analyse des Systèmes, CNRS
Open multiagent systems are systems populated with autonomous agents whose intentions are unknown. Due to this uncertainty, reputation mechanisms arise as a key technology when designing such systems. These mechanisms endow agents with the capability to reason about the behaviour of their potential partners regarding certain criteria, for instance, a particular norm. Although normative systems have been deeply studied, few attention has been paid on how agents use the norms to reason about the behaviour of their partners. In this paper, we face this problem by extending a BDI architecture that incorporates a reputation model (BDI+Repage) with a normative layer. Using the reputation mechanism together with this normative layer allow the agents to evaluate the behaviour of their partners according to both organisational and individual norms and use such information to reason about their future interactions.
Cultural Algorithms, by imitating the social behavior of human communities are able to attain good results while exhibiting a remarkable frugality in resources. This paper explores their needs in terms of sample size and how these needs are affected by the complexity of the information being addressed, with a somewhat surprising result: less is more, smaller sample sizes conduce to better results at any level of complexity.
In this paper we introduce an agent architecture for joint action negotiation among several agents in complex environments and with negotiation time bounds. The architecture is based on a graded BDI model and on an information-based negotiation model. This work does not yet include experimental results but it is currently being tested using DipGame (http://www.dipgame.org).
In complex natural systems (such as ecological systems) it is common for those studying the macro-dynamics of systems to aim to describe, as simply as possible, the interactions between individuals occupying different niches in a system and to validate these by studying whether they account for the complex network of relationships observed in the actual system. We explore a similar principle applied to multi-agent systems: we develop interactions among intelligent agents; these interactions produce complex networks of relationships among the entities in the system in a similar fashion in which ecological networks are formed in nature. The study of the topological features displayed by these networks is helpful for determining the stability of the system of software agents and its relation to the complexity of the network itself. An interaction centred approach for coordination and knowledge sharing among artificial agents is adopted for the implementation of the system.
Interactions within Multi-agent systems can be structured in different ways depending on the application scenario and its environmental restrictions. In previous work we have developed a multi-agent system within a Peer-to-Peer (P2P) sharing data network scenario. Its design includes an organisation that structures agents' interactions and an abstract architecture (2-LAMA) that helps to improve its performance. The focus of this paper is to first characterise the environment of this scenario in terms of network topologies and, secondly, to study how it affects system's performance in relation to the proposed architecture. In order to do that, we have set up a series of experiments that consider different network topologies and evaluate their performance. Results show that our architecture effectively helps to improve system's performance in despite of physical network topology.
Argumentation can be used by a group of agents to discuss about the validity of hypotheses. In this paper we propose an argumentation-based frame-work for multiagent induction, where two agents learn separately from individual training sets, and then engage in an argumentation process in order to converge to a common hypothesis about the data. The result is a multiagent induction strategy in which the agents minimize the set of examples that they have to exchange (using argumentation) in order to converge to a shared hypothesis. The proposed strategy works for any induction algorithm which expresses the hypothesis as a set of rules. We show that the strategy converges to a hypothesis indistinguishable in training set accuracy from that learned by a centralized strategy.
The implementation of AI in commercial games is usually based on low level designs that makes the control predictable, unadaptive, and non reusable. Recent algorithms such as HTN or GOAP prove that higher levels of abstraction can be applied for better performance. We propose that approaches based on Organizational Theory can help providing a sound alternative for these implementations. In this paper we present cOncienS, an integration of the ALIVE organizational framework into commercial games. We introduce a proof-of-concept implementation based on the integration to Warcraft III.
Assistive technologies represent a recent application area of a wide variety of Artificial Intelligence methods and tools to support people in their activities of daily living. But most approaches do only center in the direct interaction between the user and the assistive tool, without taking into consideration the important role that other actors (caregivers, relatives) may have in the user activities, nor they explicitly reflect the norms and regulations that apply in such scenarios. In this paper we present an approach to the development of assistive technologies which uses organisational and normative elements to ease the design of both the social network arround the user and their expected behavioural patterns.
Open Multiagent Systems, in which heterogeneous agents interact with each other and organize themselves into Virtual Organizations, demand infrastructures supporting these features. In these systems, dynamic and complex interactions between agents may arise. Interaction Protocols allow the definition of communication patterns. However in open systems, dynamic and complex interactions may also require these patterns be modified at execution time. We propose a support for modelling complex, concurrent and dynamic interactions between agents in terms of conversations. Conversations between agents follow predefined Interaction Protocols that can be dynamically modified without restarting the system. This support is provided at agent level and is integrated into the Magentix Multiagent Platform.
The use of software agents in mental health is a relatively few explored field, but that in the last years has attracted the interest of researchers due to recent studies that show the effectiveness of computarised psychological therapies. The treatment of Major Depression is one of the mental health treatments that needs the active participation, during all their phases, of the people affected by this illness. In this paper we propose the development of a novel computer-based system to support the treatment of major depression by the remote monitoring of patients and the promotion of healthy behaviours, through a Virtual Agent (VA), in response to monitored inputs. We describe the general ideas and the first steps done towards the development of the three main components in the system. We particularly emphasise the description of the Virtual Agent, which will act as the virtual peer of the patient supporting him/her with specific activities in the treatment that would contribute to an earlier return to normal health and social and economic activity.
Vessel maintenance entails periodic visual inspections of internal and external parts of the vessel hull in order to detect cracks and corroded areas. Typically, this is done by trained surveyors at great cost. Clearly, assisting them during the inspection process by means of a fleet of robots capable of defect detection would decrease the inspection cost. In this paper, two algorithms are presented for visual detection of the aforementioned two kinds of defects. On the one hand, the crack detector is based on a percolation process that exploits the morphological properties of cracks in steel surfaces. On the other hand, the corrosion detector follows a supervised classification approach taking profit from the spatial distribution of color in rusty areas. Both algorithms have shown successful rates of detection with close to real-time performance.
Complex Event Processing (CEP) consists in processing events happening across all the layers of an organization, identifying significant ones or meaningful combinations of those events and launching managing rules as a reaction. In this work we propose to use Complex Event Processing (CEP) as a supervision method for bike hiring services. The idea is to use CEP as a monitoring strategy. The events are generated by the system when users pick up or leave bikes from /to the depots distributed across the city. CEP allows describing the behaviour of the overall cycling system as a set of events, complex events and rules. It can monitor users' behaviour and trigger alarms in order to apply the corresponding billing penalties. Our CEP prototype has been tested using data extracted from the Barcelona cycling (Bicing) system. The preliminary results show he benefits of using such technology for monitoring purposes in distributed environments as the public cycling transport.
This paper surveys the exploration of the shared autonomy concept on the context of assistive technologies, in particular using an Intelligent Tutoring Ser vice to support the performance of activities of daily living while reinforcing a person's intrinsic abilities and relieving caregivers from full time assistance. The key feature of this tutoring system is its capability to adapt the service to the user's medical profile and his/her environmental context. We present the obtained outcomes while designing and testing the service on diverse scenarios with real elder volunteers having different disability profiles.
The paper presents a recommender system that permits to manage user preferences using linguistic criteria and, after collecting information about selections made by the user, it performs an unsupervised adaptation of the user profile. It has been implemented as a Web application and designed in a generic way so that it can be applied to any decision making problem. It includes two separate modules: a module to rate and rank all alternatives received by the system according to the current interests of the user, and a module to adapt the current user profile in an unsupervised fashion collecting implicit information about the user interaction with the system. The paper presents some preliminary results and discusses the performance of the adaptation algorithm.
The exploitation of sensible data associated to individuals requires a proper anonymization in order to preserve the privacy. Even though several masking methods have been designed for numerical data, very few of them deal with textual information. During the masking process, information loss should be minimized in order to enable a proper analysis of data with data mining methods. In the case of textual data, the quality of the anonymized dataset is closely related to the preservation of semantics, a dimension which has been only shallowly considered in some previous works, by using small and ad-hoc hierarchies of words. In this work we want to study the use of large and standard ontologies as the base to perform the anonymization of textual variables. We will evaluate the role of ontologies in preserving the utility of the anonymized information when a partition of the objects is done with unsupervised clustering methods. Results show that by exploiting detailed ontologies, one is able to improve the preservation of the data semantics in comparison to approaches based on ad-hoc structures and data distribution metrics.
Recommendation systems leverage product and community information to target products to consumers. Researchers have developed collaborative recommendation systems, content-based recommendation systems and a few hybrid systems. We propose a semantic framework to overcome common limitations of current systems. We present a system whose representations of items and user-profiles are based on concept taxonomies in order to provide personalized recommendation and services. The recommender incorporates semantics to enhance (1) user modeling by applying a domain-based inference method, and (2) recommendation by applying a semantic-similarity method. We show that semantics can often be used to overcome information scarcity. Experiments on movie-data from Netflix show that systems incorporating semantics produce significantly better quality recommendations than content-based ones.
This paper presents the foundation for a new methodology for a collaborative recommender system (RS). This methodology is based on the degree of consensus of a group of users stating their preferences via qualitative orders-of-magnitude. The structure of distributive lattice is considered in defining the distance between users and the RSs new users. This proposed methodology incorporates incomplete or partial knowledge into the recommendation process using qualitative reasoning techniques to obtain consensus of its users for recommendations.
In this work, we present a new two-stage technique to find clusters of different shapes, densities and sizes in the presence of overlapped clusters and noise. Firstly, a density-based clustering approach is developed using a density function estimated by the EM algorithm and in the second stage, a hierarchical strategy is used to merge clusters according to a dissimilarity measure here introduced in order to assess the overlap and proximity of the clusters. Several synthetic and real world data sets are used to evaluate the effectiveness and the efficiency of the new algorithm, indicating that it obtains satisfactory clustering results.