ARTIFICIAL INTELLIGENCE (5)

Author: RUIZ AGUILERA DANIEL.
Year: 2006.
University: ISLAS BALEARES [www.uib.es].
Place of defense: ESCUELA POLITÉCNICA SUPERIOR.
Place of preparation: ESCUELA POLITÉCNICA SUPERIOR.

Summary: The uninormas are some aggregation operators who, by their definition can be regarded as conjunctions or disyunciones and that have been applied to many different fields. This paper explores some functional equations are unknowns the uninormas or operators defined from them. One is the distributividad, which solves for classes uninormas known, solving, in particular, an open problem in the theory of non-standard analysis. It also discusses the implications and strong residual defined after uninormas, finding solution to the distributividad these implications on uninormas. As implementation of these studies, revises and expands the mathematical morphology blurred based uninormas, which provides an initial framework conducive to a new approach in the analysis of images, which will be studied in more depth.

Author: INGOLOTTI HETTER LAURA PAOLA.
Year: 2006.
University: POLITÉCNICA DE VALENCIA [www.upv.es].
Place of defense: Dep. Sistemas Informaticos y Computacion.
Place of preparation: Universidad Politécnica de Valencia.

Summary: The problems of optimization and constraint satisfaction are extremely complex and varied. At the same time, problems are of great interest, both in appearance and in the scientific and technical implemented. Therefore, to have efficient and flexible algorithmic solutions requires a high added value in a wide variety of application environments. Among the most typical problems of this kind, there are the problems of scheduling, involving the execution of actions that require resources whose availability is limited and therefore should be allocated in an efficient manner. Problems of this type appear in many contexts, including the generation of optimized schedules. Specifically, the problem considered in this thesis lies with the generation of optimized schedules rail. This problem can be likened to a problem of scheduling type Job Shop, albeit with some specificity and especially complex. Many works have been published in relation to the problem of scheduling Job Shop generic type, most of them modeling the problem through mixed integer programming and resolviéndolo using mathematical methods. In other cases, it has been considered the application of techniques CSPs hybrid, which force the consistency of restrictions up to a certain level, combining the procedure with heuristic search. However, in all papers reviewed is considered a small subset of the restrictions of the problem that we believe in this thesis. This clearly reduces the complexity of the problem, which is why in mathematical methods can be used relaxation Lagrangiana or cutting plans to reduce the search space. However, when entering the complex types of restrictions that appear in real contexts, the mathematical programming is no longer viable, because the solution of mathematical methods become impractical.

Author: BRITO RODRIGUEZ ISMEL.
Year: 2006.
University: POLITÉCNICA DE CATALUÑA [www.upc.edu].
Place of defense: Sala d'Actes. Facultat d'Informàtica.
Place of preparation: OMEGA DESPATX 240 ND.

Summary: In recent years, the Artificial Intelligence community has shown an increasing interest in solving distributed problems using the agents paradigm. When multiple agents in a shared environment pursue a common goal, there are usually constraints among the possible actions of these agents. Finding a consistent combination of actions that satisfies agents’ constraints can be seen as a Distributed Constraint Satisfaction problem (DisCSP). Various application problems in multi-agent systems can be formalized as DisCSPs. This thesis is dedicated to the development of distributed complete algorithms for solving DisCSP. In it, we study three types of algorithms: synchronous, asynchronous and hybrid. We evaluate the proposed algorithms in two dimensions: efficiency and privacy. Regarding efficiency, we propose new distributed algorithms which mainly are faster and consume less network resources than state-of-the-art algorithms. Regarding privacy, we propose novel algorithms to enforce the privacy of the local information held by agents without using cryptographic tools. The main ideas that we have developed in this thesis are: Synchronous Algorithms: The use of variable reordering heuristics for constraint satisfaction problems has been shown to be a powerful strategy in order to improve efficiency. Inspired in this idea, we present two approximations of the popular minimumdomain heuristic for dynamic variable reordering. Asynchronous Algorithms: We present a basic kernel for grouping asynchronous backtracking algorithms. By implementing the condition for termination in this kernel, we obtain four asynchronous algorithms. One of these algorithms does not add links between agents not sharing constraints, which can be useful for solving problems where privacy is the main concern. Hybrid Algorithms: We present a novel algorithm which combines synchronous and asynchronous elements. This algorithm outperforms the reference asynchronous algorithm. Non-binary Constraints: Although most of state-of-the-art methods for DisCSP assume that every constraint involves two variables, they can be extended to handle constraints involving more than two variables. We present new versions of existing algorithms to deal with non-binary constraints, including the addition of redundant constraint projections. Assignment Privacy: We propose an asynchronous algorithm that allows agents to maintain their variable assignments private during problem resolution. Constrain Privacy: We present the Partially Known Constraint model (PKC), a new DisCSP model in which constraints are kept private and are only partially known to agents. We propose two algorithms for solving DisCSP expressed under the PKC model. Both algorithms also consider assignment privacy. Enforcing Privacy with Lies: We present a novel algorithm that further enforces constraint privacy. This is algorithm is based on the idea that agents may lie. It requires a single extra condition: if an agent lies, it has to tell the truth in finite time afterwards. Applications: We consider naturally distributed constraint problems which have a clear motivation to be tried with distributed techniques: Meeting Scheduling and Stable Matching problems. For each these problems we present distributed versions. Regarding Stable Matching problems, we consider two well-known problems: Stable Marriage and Stable Roommates problems. We propose ways to solve these problems while keeping personal preferences private. All proposed algorithms in this thesis have been implemented, evaluated and formally proven to be correct, complete and terminate.

Author: MORALES FERNANDEZ EMILIO JESUS.
Year: 2006.
University: CÓRDOBA [www.uco.es].
Place of defense: ETEA.
Place of preparation: ETEA.

Summary: The centerpiece of this work is ESTRUCTURACIÓN OF A MODEL THEORY ON RELATIONS that, based on theoretical research and empirical existing permit COMBINE DESIGN CONTINGENTE OF THE ORGANIZATIONAL STRUCTURE AND THE STRATEGY TO PROPOSE BUSINESS PRACTICES OF HUMAN RESOURCES MANAGEMENT (HRM ) APPROPRIATE. Building on this theoretical model, IS DESIGNED AND BUILT AN EXPERT SYSTEM (SE) ANALYSIS FOR THE INTEGRATED of systems considered -estructural, strategic and HRM. First, the adjustments have been identified and existing maladjustments both internally and with the systems that are analyzed, secondly, it has facilitated prescribing policies and practices of HRM, and thirdly, it provides a instrumetno work - the SE - which is useful in the field of teaching and the investigaicón. THE PROCESS METODOLÓGICO Followed consists of four phases: 1-A literature review of the most significant contributions as restrained and empirical knowledge in the areas concerned, as well as the definition of a comprehensive model of relationships that incorporates existing relationships between the basic variables . 2 - The search, definition and development of tools for measuring variables of the model defined above. 3 - The design and construction of a SE assumes that the relations of the theoretical model, which facilitates the analysis, drawing conclusions, and suggested alternative design and organizational change, based on actual specific cases. 4-Testing and evaluation of the validity of SE from real specific cases. The outcome of the work has been organized into two parties, one theoretical and practical. The PART THEORETICAL are the chapters 1, 2 and 3 and was told to review, analyze and choose from theoretical and empirical frameworks of the systems studied, aspects needed to generate the model of relations. PART PRACTICAL develops in Chapters 4, 5 and 6, establishing connections between the elements and selected aspects in the theoretical part by a MODEL OF THE RELATIONSHIP. This is the basis on which the software generates SYSTEM EXPERT CEPRU 1.0-capable of applying the theoretical model and analysis expert who provides information on human resources practices most appropriate Once built is the result of A SIMULATION CASES OF COMPANIES WITH REAL analyzing the exits and indications offered by the program in each, to measure the validity of it.

Author: JUAREZ HERRERO JOSE MANUEL.
Year: 2006.
University: MURCIA [www.um.es].
Place of defense: FACULTAD DE INFORMÁTICA.
Place of preparation: FACULTAD DE INFORMATICA.

Summary: In the area of intensive care medicine, the actual implementation of support systems for the diagnosis requires a representation and proper management of diagnostic knowledge. However, in domains highly complex concept, where the temporal dimension plays a key role, knowledge diagnosis is not always affordable for a single approximation of the Artificial Intelligence. So, in summary of this thesis deals with the issue of temporary diagnosis through different modes of reasoning (Case-Based Reasoning, Reasoning Based Reasoning Models and Multimode) and its actual implementation in an Intensive Care Unit, summarizing some of the outcomes of the doctoral thesis. 1. MOTIVATION Success in the actual use of systems to aid in medical diagnosis depends partly on the ability to model and manage the knowledge domain and the temporal dimension. A growing trend is to propose theoretical models where it says (or easier) to the domain model that is applicable. Despite the reduction and simplification of domains is effective in specific diagnostic problems, interdisciplinary certain fields of medicine, such as the Intensive Care Unit (ICU), involving a highly complex concept, because of the high number of concepts, interrelationships, terminology among its elements and temporality. Thus, we believe that, in domains such as the ICU, it is necessary to address the problem of temporary diagnosis from a generic approach of modeling knowledge (eg description of pathology) and from the knowledge of each particular patient. To this end, we study both the Models Based Reasoning (MBR), in Section 2, as in Case Based Reasoning (CBR), in Section 3, demonstrating its applicability by developing prototypes validated in the domain of the UCI. In addition, we will analyze how these two modes of reasoning can be combined to improve the experimental results (Section 4), called Reasoning Multimode (MMR). 2. RAZONAMIENTO BASED MODELS The diagnostic techniques based models have proved to be suitable for modeling of knowledge and reasoning in diagnostic medical domains. In this approach we propose a model of causal / time, a method of diagnosis and architecture applied in medical domains. Thus, we propose a first Model Behavior Temporary (TBM), which describe the abnormal behavior of patients, represented relations between the abnormal findings and disease. This model breaks down into Diagnosis Seasonal Patterns Borrosos or dftp, where every employer establishes the causal relationships between an etiologic diagnosis and its implications (findings and diagnostic patofisiológicos). The relationship between these elements are causal relationships and timing relationships vague, represented by the networks Restrictions Temporary Borrosas (FTCN) proposals. In addition, each pattern includes temporary knowledge of the context, which affects the temporal evolution of a disease. This context is called Metaconocimiento Contextual Temporal, whose elements (elements of context) also can define timing relationships. For example, in the medical domain, the increase in blood pressure of a patient (when it is too low) because the appropriate treatment could establish a contextual factor to be taken into account when interpreting the measurement result of pressure from patient . The task diagnostic study must obtain a causal explanation (diagnostic hypothesis) from a set of observations temporary changes in the patient (temporary problem diagnosis). Thus, we propose a method of diagnosis consists of two steps: (1) construction of form 8 to abduct 1ee5 vat of a causal network (consisting temporarily), taking advantage of the findings observed; (2) eliminate those elements of the network through causal a method of discrimination hypothesis. With in order to demonstrate the applicability of TBM model, we have developed ACUDES, an architecture for decision making in ICU. ACUDES manages representation of the domain of application using an ontology for ICU providing consistency conceptual architecture. The knowledge base system contains a set of temporal patterns collected through a diagnostic tool for acquiring knowledge. The clinical evaluation of ACUDES have included construction of a model TBM full descriptions in a real pathological ICU by a doctor intensivista. The evaluation can be summarized in terms of: 3 sessions spot with the expert, TBM obtained (5 patterns temporary), ontology ICU (173 concepts and 105 relations) and resolved correctly 3 actual cases studied. 4. RAZONAMIENTO BASED ON CASE The resolution of problems through analog processes is particularly appropriate domains loosely structured and where the acquisition of knowledge is difficult. Thus, the CBR is a methodology that in the past two decades has shown itself capable of resolving problems in the medical field. Thus, we believe that the proposal for a diagnostic system CBR in a domain highly complex concept must focus on the recovery of similar cases temporarily. Thus, if a patient is composed by a set of timeless characteristics (eg family history or demographic data) and temporal (when tested for image or length of treatment). Unlike the timeless characteristics, and sets up temporary sequences values are difficult to compare. A subset usual problems are those where all its attributes are references to moments of time (time points), describing temporal sequence of events. For example, a unit of Great Burns, developments in the study of parameters such as pH of the skin or diuresis at different times. In order to solve this problem, we have proposed a measure of similarity of temporal sequence of events, based on the networks Restrictions Posibilísticas (PTCN), using two steps: (1) isolate two or more sequences in a time frame of possible links temporary between the characteristics (2) quantify the uncertainty caused by the description posibilística through non-specificity (a measure of uncertainty), as this is an indicator of the similarity of the sequences. However, it is common to find descriptions on temporary intervals. For example, the description of the treatment of a patient is described by the set of drugs provided to a patient at length, describing sequences intervals. Thus, we propose a measure of similarity between sequences intervals recital: direct comparison between intervals, comparing relations between temporary qualitative / quantitative and the presence / absence of intervals respect to the other stream. For comparison between comprehensive information temporary sequence intervals, this measure uses a network of temporary restrictions, qualitative and quantitative, between intervals (ABIN). As a practical implementation techniques similar proposal we have developed T-CARE, a system for decision-making in the domain of the UCI. The objective of T-CARE is recovering patient cases similar to aid diagnosis, searching in a library Cases (LC) patients its temporal evolution is similar. Thus, the system T-CARE consists of two processes: (1) acquisition of temporary cases from clinical information system, and (2) recovery of similar cases. To facilitate the process of acquisition of cases from the clinical information systems (HIS), we have implemented a set of visualization tools temporary medical histories and assistants for selecting which elements of the patient's history are part of the case. The system kernel T-CARE is the module recovery cases. The process of recovery cases, it will look at the LC, most similar cases temporarily. T-CARE provides various filters (comprehensiveness of the case and quality of the event) to reduce the search space, then applying the measures of similarity based on similarity of temporal sequences of events and intervals. Validation System T-CARE has been to solve a clinical problem for the Big Unit Burns (UGQ) by implementing conducted T-CARE. In UGQ, clinical studies demonstrate the survival of patients after its temporal evolution during the first 5 days. 5. RAZONAMIENTO MULTIMODAL In order to improve the experimental results obtained in the MBR and CBR approaches, we considered architectures that combine these two strategies together, called multimodal. While the approximation based on models allow formalize the general knowledge, the approximation based on analogy allow exploit the more specific aspects. In this work we have proposed and developed M ^ 3, a multimodal architecture that uses CBR and MBR to solve the problem of temporary diagnosis. The knowledge base of M ^ 3 can be used to store both cases temporal patterns as diagnostics. Thus, the method of diagnosis can exploit a broader spectrum of diagnostic knowledge. The proposed method of diagnosis automatically selects one of two modes of reasoning or both combined. This method depends on a parameter configurable by the expert who sets the threshold of similarity, establishing whether a case is similar to another or not. The clinical validation of M ^ 3 that has been done in the ICU for patients with heart disease indicates that an appropriate configuration of this parameter, in addition to improving the rate hits diagnostics, significantly improves the specificity and sensitivity regarding T-CARE and ACUDES . CONCLUSIONS AND FUTURE WORK In this paper we have addressed the problem of diagnosis temporary domains highly complex concept, proposing a set of techniques and architectures for the three approximations considered (MBR CBR and MMR) and its validation in the field of medicine. The combination of techniques in the MMR has proved in the experiments, the approach that best results have been achieved in the domain of the UCI. The modeling of time depending on the purpose (consistency, etc. through temporary measure similarity models FTCN, PTCN and ABIN, has proved to be a suitable approximation to manage the temporal dimension. In future work will attempt to improve different aspects of technical proposals (integration patterns in the MBR and adaptation CBR), as well as its assessment in other areas of medicine.

Author: GARCÍA SÁNCHEZ FRANCISCO.
Year: 2006.
University: MURCIA [www.um.es].
Place of defense: FACULTAD DE INFORMÁTICA.
Place of preparation: FACULTAD DE INFORMÁTICA, UNIVERSIDAD DE MURCIA.

Summary: In this dissertation, it is the fact that, conceptually, intelligent agents and Web Services (semantic) were conceived with totally disparate purposes and, as such, it is understood that (scenario) should stay at two different levels of abstraction. The main idea that agents based technology is not that intelligent agents are able to provide services, but those are designed as autonomous entities that incorporate intelligence and cognitive abilities that enable them to display a proactive behavior-oriented goals and establish processes interaction, whether competitive or cooperative, with other entities to meet their design goals. For its part, Web services represented a new trend in the world of distributed computing, and have the sole purpose of providing functionality from software components globally accessible. Such conceptual differences justify the need for both technologies in an integrated environment and provide clues about the benefits to be gained if applied to tasks with different levels of abstraction for developing complex systems. Based on this assumption departure, in this thesis has developed a framework that makes use of the agents and Web services Semánticos for developing applications that can deal with the dynamism of the Web, while also they can benefit from features such as self-reliance, learning and reasoning. This is the point at which cobra relevancia Engineering Ontológica. The ontologies are the components that allow communication between agents and Web Services, located at different levels of abstraction occurs in a smooth fashion and without misinterpretation. The architecture of the working framework developed consists essentially of a multi-environment, a set of knowledge bases and various interfaces that allow the system to communicate them effectively with external entities identified, namely Web Services and Suppliers service entities (users) consumer services, and developers.