This thesis is dedicated to the development of statistical and computational methods for the analysis of DNA sequences and gene expression time series.<br /><br />First we study a parsimonious Markov model called Mixture Transition Distribution (MTD) model which is a mixture of Markovian transitions. The overly high number of constraints on the parameters of this model hampers the formulation of an analytical expression of the Maximum Likelihood Estimate (MLE). We propose to approach the MLE thanks to an EM algorithm. After comparing the performance of this algorithm to results from the litterature, we use it to evaluate the relevance of MTD modeling for bacteria DNA coding sequences in comparison with standard Markovian modeling.<br /><br />Then we propose two different approaches for genetic regulation network recovering. We model those genetic networks with Dynamic Bayesian Networks (DBNs) whose edges describe the dependency relationships between time-delayed genes expression. The aim is to estimate the topology of this graph despite the overly low number of repeated measurements compared with the number of observed genes. <br /><br />To face this problem of dimension, we first assume that the dependency relationships are homogeneous, that is the graph topology is constant across time. Then we propose to approximate this graph by considering partial order dependencies. The concept of partial order dependence graphs, already introduced for static and non directed graphs, is adapted and characterized for DBNs using the theory of graphical models. From these results, we develop a deterministic procedure for DBNs inference. <br /><br />Finally, we relax the homogeneity assumption by considering the succession of several homogeneous phases. We consider a multiple changepoint<br />regression model. Each changepoint indicates a change in the regression model parameters, which corresponds to the way an expression level depends on the others. Using reversible jump MCMC methods, we develop a stochastic algorithm which allows to simultaneously infer the changepoints location and the structure of the network within the phases delimited by the changepoints. <br /><br />Validation of those two approaches is carried out on both simulated and real data analysis.

<p>Vid analys av fluider med akustisk spektroskopi finns ett behov av att finna multivariata metoder för att utifrån akustiska spektra prediktera storheter såsom viskositet och densitet. Användning av artificiella neurala nätverk och bayesiska nätverk för detta syfte utreds genom teoretiska och praktiska undersökningar. Förbehandling och uppdelning av data samt en handfull linjära och olinjära multivariata analysmetoder beskrivs och implementeras. Prediktionsfelen för de olika metoderna jämförs och PLS (Partial Least Squares) framstår som den starkaste kandidaten för att prediktera de sökta storheterna.</p> / <p>When analyzing fluids using acoustic spectrometry there is a need of finding multivariate methods for predicting properties such as viscosity and density from acoustic spectra. The utilization of artificial neural networks and Bayesian networks for this purpose is analyzed through theoretical and practical investigations. Preprocessing and division of data along with a handful of linear and non-linear multivariate methods of analysis are described and implemented. The errors of prediction for the different methods are compared and PLS (Partial Least Squares) appear to be the strongest candidate for predicting the sought-after properties.</p>

Graphical techniques for modeling the dependencies of randomvariables have been explored in a variety of different areas includingstatistics, statistical physics, artificial intelligence, speech recognition, image processing, and genetics.Formalisms for manipulating these models have been developedrelatively independently in these research communities. In this paper weexplore hidden Markov models (HMMs) and related structures within the general framework of probabilistic independencenetworks (PINs). The paper contains a self-contained review of the basic principles of PINs.It is shown that the well-known forward-backward (F-B) and Viterbialgorithms for HMMs are special cases of more general inference algorithms forarbitrary PINs. Furthermore, the existence of inference and estimationalgorithms for more general graphical models provides a set of analysistools for HMM practitioners who wish to explore a richer class of HMMstructures.Examples of relatively complex models to handle sensorfusion and coarticulationin speech recognitionare introduced and treated within the graphical model framework toillustrate the advantages of the general approach.

Decision-theoretic systems, such as Markov Decision Processes (MDPs), are used for sequential decision-making under uncertainty. MDPs provide a generic framework that can be applied in various domains to compute optimal policies. This thesis presents techniques that offer explanations of optimal policies for MDPs and then refine decision theoretic models (Bayesian networks and MDPs) based on feedback from experts. Explaining policies for sequential decision-making problems is difficult due to the presence of stochastic effects, multiple possibly competing objectives and long-range effects of actions. However, explanations are needed to assist experts in validating that the policy is correct and to help users in developing trust in the choices recommended by the policy. A set of domain-independent templates to justify a policy recommendation is presented along with a process to identify the minimum possible number of templates that need to be populated to completely justify the policy. The rejection of an explanation by a domain expert indicates a deficiency in the model which led to the generation of the rejected policy. Techniques to refine the model parameters such that the optimal policy calculated using the refined parameters would conform with the expert feedback are presented in this thesis. The expert feedback is translated into constraints on the model parameters that are used during refinement. These constraints are non-convex for both Bayesian networks and MDPs. For Bayesian networks, the refinement approach is based on Gibbs sampling and stochastic hill climbing, and it learns a model that obeys expert constraints. For MDPs, the parameter space is partitioned such that alternating linear optimization can be applied to learn model parameters that lead to a policy in accordance with expert feedback. In practice, the state space of MDPs can often be very large, which can be an issue for real-world problems. Factored MDPs are often used to deal with this issue. In Factored MDPs, state variables represent the state space and dynamic Bayesian networks model the transition functions. This helps to avoid the exponential growth in the state space associated with large and complex problems. The approaches for explanation and refinement presented in this thesis are also extended for the factored case to demonstrate their use in real-world applications. The domains of course advising to undergraduate students, assisted hand-washing for people with dementia and diagnostics for manufacturing are used to present empirical evaluations.

For the manufacture of gypsum powder applied and checked on its setting, using multiple components. One of the intrinsic components of construction gypsum is the anhydrite, completely dehydrated gypsum, working as inert part. Bayesian networks are one of the tools of industrial management. One of its most important properties is its capacity to self-learning. This research work builds on the improvement of anhydrite calcination process using a rotary kiln direct cooking. The work provide an update as new systems currently used for the manufacture of anhydrite, since in addition to improving process control via control loops more efficient and self-managed, and improvements in levels MES and SCADA provide artificial intelligence elements by applying the above Bayesian networks. / En la fabricación de yeso en polvo, aplicable y controlable en su endurecimiento, se utilizan varios componentes. Uno de los componentes intrínseco del yeso para construcción es la anhidrita o yeso totalmente deshidratado que trabaja como parte inerte. Las redes bayesianas, como sistema experto, son una herramienta de gestión industrial. Una de sus propiedades más importantes es su capacidad de autoaprendizaje. Esta investigación se basará en la mejora del proceso de calcinación de anhidrita usando un horno rotativo de cocción directa. El trabajo aportará como novedad una actualización de los sistemas usados actualmente para la fabricación de anhidrita, ya que aparte de mejorar el control del proceso mediante lazos de control más eficientes y autogestionados, así como la introducción de mejoras en los niveles MES y Scada, aportará una modelización del proceso con elementos de inteligencia artificial mediante la aplicación de dichas redes bayesianas.

Vid analys av fluider med akustisk spektroskopi finns ett behov av att finna multivariata metoder för att utifrån akustiska spektra prediktera storheter såsom viskositet och densitet. Användning av artificiella neurala nätverk och bayesiska nätverk för detta syfte utreds genom teoretiska och praktiska undersökningar. Förbehandling och uppdelning av data samt en handfull linjära och olinjära multivariata analysmetoder beskrivs och implementeras. Prediktionsfelen för de olika metoderna jämförs och PLS (Partial Least Squares) framstår som den starkaste kandidaten för att prediktera de sökta storheterna. / When analyzing fluids using acoustic spectrometry there is a need of finding multivariate methods for predicting properties such as viscosity and density from acoustic spectra. The utilization of artificial neural networks and Bayesian networks for this purpose is analyzed through theoretical and practical investigations. Preprocessing and division of data along with a handful of linear and non-linear multivariate methods of analysis are described and implemented. The errors of prediction for the different methods are compared and PLS (Partial Least Squares) appear to be the strongest candidate for predicting the sought-after properties.

The share of peer-to-peer (P2P) protocol in the total network traffic grows dayby- day in the Turkish Academic Network (UlakNet) similar to the other networks in the world. This growth is mostly because of the popularity of the shared content and the great enhancement in the P2P protocol since it first came out with Napster. The shared files are generally both large and copyrighted. Motivated by the problems of UlakNet with the P2P traffic, we propose a novel method for P2P traffic detection in the network backbone in this thesis. Observing the similarity between detecting traffic that belongs to a specific protocol and detecting an intrusion in a computer system, we adopt an Intrusion Detection System (IDS) technique to detect P2P traffic. Our method is a passive detection procedure that uses traffic flows gathered from border routers. Hence, it is scalable and does not have the problems of other approaches that rely on packet payload data or transport layer ports.

Σκοπός της διατριβής είναι η ανάλυση, η μελέτη και η μοντελοποίηση της συμπεριφοράς τόσο των ευφυών πρακτόρων όσο και των χρηστών σε εικονικά περιβάλλοντα μάθησης, με τη χρήση τεχνικών υπολογιστικής νοημοσύνης. Το θεματικό αντικείμενο της διδακτορικής διατριβής αποτελεί ένα σύγχρονο αντικείμενο βασικής έρευνας με μεγάλο εύρος πρακτικών εφαρμογών. Η βάση της ερευνητικής δραστηριότητας εστιάζεται σε δύο βασικούς τομείς: 1. Προσαρμόσιμη μοντελοποίηση συμπεριφορών ευφυών πρακτόρων σε εικονικά περιβάλλοντα μάθησης, σύμφωνα με κανόνες βελτιστοποίησης της μαθησιακής επίδρασης στο χρήστη μέσα στο εικονικό περιβάλλον μάθησης. 2. Μοντελοποίηση χρηστών εικονικών περιβαλλόντων μάθησης, με στόχο τη βελτιστοποίηση της μαθησιακής επίδρασης στο χρήστη. Για τη μοντελοποίηση, τόσο της συμπεριφοράς των ευφυών πρακτόρων, όσο και των χρηστών, χρησιμοποιήθηκαν προηγμένες τεχνικές υπολογιστικής νοημοσύνης (Bayesian Δίκτυα, Γενετικοί και Εξελικτικοί Αλγόριθμοι). Αυτές οι τεχνικές, εκτός από την ευφυΐα, ενσωματώνουν και το επιθυμητό χαρακτηριστικό της προσαρμοσιμότητας, με την έννοια ότι μπορούν να προσαρμόζονται στις αλλαγές του περιβάλλοντος. Τα παραπάνω αποτελέσματα αξιολογήθηκαν στη χρήση τους σε Ευφυή Εικονικά Συστήματα Μάθησης βασισμένα στο Web (Intelligent Virtual Learning Systems – IVLS), τα οποία αποτελούν ουσιαστικά το μέσον εξαγωγής συμπερασμάτων και υποστηρικτικού υλικού για τη μετρήσιμη συμπεριφορά τόσο των ευφυών πρακτόρων όσο και των χρηστών, μέσα σε τέτοια περιβάλλοντα. / The main objectives of the thesis are the analysis, study and the provision of a behavior modeling procedure of the intelligent agents and the students in virtual e-learning systems using computational intelligence techniques. The domain of the thesis is a topic of basic research with a large scale of applied results. The basis of the research is focused in two main sectors: 1. Adaptive behavior modeling of intelligent agents in virtual learning systems, according to specific optimization rules of the learning process during the interaction of the user/student with the e-learning environment. 2. User modeling of the users of virtual learning environments towards the optimization of the learning process. For the modeling procedure of the behavior of intelligent agents and of the users specific computational intelligence techniques have been applied (Bayesian Networks, Genetic και Evolutionary Algorithms). The specific techniques provide intelligence to the system and the most important the feature of adaptability. The aforementioned results have been evaluated on Intelligent Virtual Learning Systems, which constitute the medium for the inference of the results and the mean for supportive material for the measurable behavior of the intelligent agents and of the users in Intelligent Virtual Learning Systems.

Οι τεχνικές Υπολογιστικής Νοημοσύνης βρίσκουν σε μεγάλο βαθμό εφαρμογή σε Ηλεκτρονικά Συστήματα Μάθησης. Στην εργασία αυτή υιοθετείται η τεχνική των Bayesian δικτύων. Αναλυτικότερα υλοποιείται ένα έξυπνο σύστημα το οποίο αναλαμβάνει τη διαχείριση των ερωτηματολογίων ενός Ηλεκτρονικού Συστήματος Μάθησης. Σκοπός της των Bayesian δικτύων είναι η «έξυπνη» διαχείριση των ερωτηματολογίων. Πιο συγκεκριμένα, πραγματοποιείται γραφική απεικόνιση των ερωτηματολογίων σε Bayesian γράφημα όπου κάθε ερώτηση αντιστοιχεί σε ένα κόμβο του γραφήματος. Στο γράφημα αυτό εφαρμόζονται οι εξισώσεις του Bayes σε κάθε κόμβο του γραφήματος ώστε να υπολογιστούν οι πιθανότητες επιτυχούς απάντησης μιας ερώτησης. Στη συνέχεια οι πιθανότητες συγκρίνονται με κατώφλια τα οποία ορίζει ο διαχειριστής του συστήματος ώστε να αποφευχθούν ερωτήσεις στις οποίες ο χρήστης έχει μεγάλη πιθανότητα να απαντήσει επιτυχώς. Επίτευγμα αυτής της υλοποίησης είναι η εξοικονόμηση ερωτήσεων και χρόνου εκ μέρους του χρήστη. Το δεύτερο μέρος της εργασίας αφορά στην επέκταση του παραπάνω συστήματος χρησιμοποιώντας την αρχιτεκτονική ευφυών πρακτόρων. Βασικός σκοπός της επέκτασης αυτής είναι η δυνατότητα διαχείρισης ενός μεγάλου αριθμού χρηστών και ερωτηματολογίων από απομακρυσμένα συστήματα. / In this contribution an innovative platform is being presented that integrates intelligent agents in legacy e-learning environments. It introduces the design and development of a scalable and interoperable integration platform supporting various assessment agents for e-learning environments. The agents are implemented in order to provide intelligent assessment services to computational intelligent techniques such as Bayesian Networks and Genetic Algorithms. The utilization of new and emerging technologies like web services allows integrating the provided services to any web based legacy e-learning environment.

Decision-theoretic systems, such as Markov Decision Processes (MDPs), are used for sequential decision-making under uncertainty. MDPs provide a generic framework that can be applied in various domains to compute optimal policies. This thesis presents techniques that offer explanations of optimal policies for MDPs and then refine decision theoretic models (Bayesian networks and MDPs) based on feedback from experts. Explaining policies for sequential decision-making problems is difficult due to the presence of stochastic effects, multiple possibly competing objectives and long-range effects of actions. However, explanations are needed to assist experts in validating that the policy is correct and to help users in developing trust in the choices recommended by the policy. A set of domain-independent templates to justify a policy recommendation is presented along with a process to identify the minimum possible number of templates that need to be populated to completely justify the policy. The rejection of an explanation by a domain expert indicates a deficiency in the model which led to the generation of the rejected policy. Techniques to refine the model parameters such that the optimal policy calculated using the refined parameters would conform with the expert feedback are presented in this thesis. The expert feedback is translated into constraints on the model parameters that are used during refinement. These constraints are non-convex for both Bayesian networks and MDPs. For Bayesian networks, the refinement approach is based on Gibbs sampling and stochastic hill climbing, and it learns a model that obeys expert constraints. For MDPs, the parameter space is partitioned such that alternating linear optimization can be applied to learn model parameters that lead to a policy in accordance with expert feedback. In practice, the state space of MDPs can often be very large, which can be an issue for real-world problems. Factored MDPs are often used to deal with this issue. In Factored MDPs, state variables represent the state space and dynamic Bayesian networks model the transition functions. This helps to avoid the exponential growth in the state space associated with large and complex problems. The approaches for explanation and refinement presented in this thesis are also extended for the factored case to demonstrate their use in real-world applications. The domains of course advising to undergraduate students, assisted hand-washing for people with dementia and diagnostics for manufacturing are used to present empirical evaluations.