H παρούσα διατριβή εκπονήθηκε στο Εργαστήριο Αναγνώρισης Προτύπων, του Τμήματος Μηχανικών Ηλεκτρονικών Υπολογιστών και Πληροφορικής του Πανεπιστημίου Πατρών. Αποτελεί μέρος της ευρύτερης ερευνητικής δραστηριότητας του Εργαστηρίου στον τομέα του σχεδιασμού και της εφαρμογής των τεχνολογιών Υπολογιστικής Νοημοσύνης στην ανάλυση βιολογικών δεδομένων. Η διδακτορική αυτή διατριβή χρηματοδοτήθηκε από το πρόγραμμα Ηράκλειτος ΙΙ. Ο τομέας της πρωτεωμικής είναι ένα σχετικά καινούργιο και γρήγορα αναπτυσσόμενο ερευνητικό πεδίο. Μια από τις μεγαλύτερες προκλήσεις στον τομέα της πρωτεωμικής είναι η αναδόμηση του πλήρους πρωτεϊνικού αλληλεπιδραστικού δικτύου μέσα στα κύτταρα. Εξαιτίας του γεγονότος, ότι οι πρωτεϊνικές αλληλεπιδράσεις παίζουν πολύ σημαντικό ρόλο στις βασικές λειτουργίες ενός κυττάρου, η ανάλυση αυτών των δικτύων μπορεί να αποκαλύψει τον ρόλο αυτών των αλληλεπιδράσεων στις ασθένειες καθώς και τον τρόπο με τον οποίο οι τελευταίες αναπτύσσονται. Παρόλα αυτά, είναι αρκετά δύσκολο να καταγραφούν και να μελετηθούν οι πρωτεϊνικές αλληλεπιδράσεις ενός οργανισμού, καθώς το πρωτέωμα διαφοροποιείται από κύτταρο σε κύτταρο και αλλάζει συνεχώς μέσα από τις βιοχημικές του αλληλεπιδράσεις με το γονιδίωμα και το περιβάλλον. Ένας οργανισμός έχει ριζικά διαφορετική πρωτεϊνική έκφραση στα διάφορα σημεία του σώματός του, σε διαφορετικά στάδια του κύκλου ζωής του και υπό διαφορετικές περιβαλλοντικές συνθήκες. Δημιουργούνται, λοιπόν, δύο πάρα πολύ σημαντικοί τομείς έρευνας, που είναι, πρώτον, η εύρεση των πραγματικών πρωτεϊνικών αλληλεπιδράσεων ενός οργανισμού που θα συνθέσουν το πρωτεϊνικό δίκτυο αλληλεπιδράσεων και, δεύτερον, η περαιτέρω ανάλυση του πρωτεϊνικού δικτύου για εξόρυξη πληροφορίας (εύρεση πρωτεϊνικών συμπλεγμάτων, καθορισμός λειτουργίας πρωτεϊνών κτλ). Στην παρούσα διδακτορική διατριβή παρουσιάζονται καινοτόμες αλγοριθμικές τεχνικές Υπολογιστικής Νοημοσύνης για την πρόβλεψη πρωτεϊνικών αλληλεπιδράσεων, τον υπολογισμό ενός βαθμού εμπιστοσύνης για κάθε προβλεφθείσα αλληλεπίδραση, την πρόβλεψη πρωτεϊνικών συμπλόκων από δίκτυα πρωτεϊνικών αλληλεπιδράσεων και την πρόβλεψη της λειτουργίας πρωτεϊνών. Συγκεκριμένα, στο κομμάτι της πρόβλεψης και βαθμολόγησης πρωτεϊνικών αλληλεπιδράσεων αναπτύχθηκε μια πληθώρα καινοτόμων τεχνικών ταξινόμησης. Αυτές κυμαίνονται από υβριδικούς συνδυασμούς μετα-ευρετικών μεθόδων και ταξινομητών μηχανικής μάθησης, μέχρι μεθόδους γενετικού προγραμματισμού και υβριδικές μεθοδολογίες ασαφών συστημάτων. Στο κομμάτι της πρόβλεψης πρωτεϊνικών συμπλόκων υλοποιήθηκαν δύο βασικές καινοτόμες μεθοδολογίες μη επιβλεπόμενης μάθησης, οι οποίες θεωρητικά και πειραματικά ξεπερνούν τα μειονεκτήματα των υπαρχόντων αλγορίθμων. Για τις περισσότερες από αυτές τις υλοποιηθείσες μεθοδολογίες υλοποιήθηκαν φιλικές προς τον χρήστη διεπαφές. Οι περισσότερες από αυτές τις μεθοδολογίες μπορούν να χρησιμοποιηθούν και σε άλλους τομείς. Αυτό πραγματοποιήθηκε με μεγάλη επιτυχία σε προβλήματα βιοπληροφορικής όπως η πρόβλεψη microRNA γονιδίων και mRNA στόχων τους και η μοντελοποίηση - πρόβλεψη οικονομικών χρονοσειρών. Πειραματικά, η μελέτη αρχικά επικεντρώθηκε στον οργανισμό της ζύμης (Saccharomyces cerevisiae), έτσι ώστε να αξιολογηθούν οι αλγόριθμοι, που υλοποιήθηκαν και να συγκριθούν με τις υπάρχουσες αλγοριθμικές μεθοδολογίες. Στη συνέχεια, δόθηκε ιδιαίτερη έμφαση στις πρωτεΐνες του ανθρώπινου οργανισμού. Συγκεκριμένα, οι καλύτερες αλγοριθμικές τεχνικές για την ανάλυση δεδομένων πρωτεϊνικών αλληλεπιδράσεων εφαρμόστηκαν σε ένα σύνολο δεδομένων που δημιουργήθηκε για τον ανθρώπινο οργανισμό. Αυτό είχε σαν αποτέλεσμα την δημιουργία ενός πλήρους, σταθμισμένου δικτύου πρωτεϊνικών αλληλεπιδράσεων για τον άνθρωπο και την εξαγωγή των πρωτεϊνικών συμπλόκων, που υπάρχουν σε αυτό καθώς και τον λειτουργικό χαρακτηρισμό πολλών αχαρακτήριστων πρωτεϊνών. Τα αποτελέσματα της ανάλυσης των δεδομένων πρωτεϊνικών αλληλεπιδράσεων για τον άνθρωπο είναι διαθέσιμα μέσω μίας διαδικτυακής βάσης γνώσης HINT-KB (http://hintkb.ceid.upatras.gr), που υλοποιήθηκε στα πλαίσια αυτής της διδακτορικής διατριβής. Σε αυτή την βάση γνώσης ενσωματώνεται, από διάφορες πηγές, ακολουθιακή, δομική και λειτουργική πληροφορία για ένα τεράστιο πλήθος ζευγών πρωτεϊνών του ανθρώπινου οργανισμού. Επίσης, οι χρήστες μπορούν να έχουν προσβαση στις προβλεφθείσες πρωτεϊνικές αλληλεπιδράσεις και στον βαθμό εμπιστοσύνης τους. Τέλος, παρέχονται εργαλεία οπτικοποίησης του δικτύου πρωτεϊνικών αλληλεπιδράσεων, αλλά και εργαλεία ανάκτησης των πρωτεϊνικών συμπλόκων που υπάρχουν σε αυτό και της λειτουργίας πρωτεϊνών και συμπλόκων. Το προβλήματα με τα οποία καταπιάνεται η παρούσα διδακτορική διατριβή έχουν σημαντικό ερευνητικό ενδιαφέρον, όπως τεκμηριώνεται και από την παρατιθέμενη στη διατριβή εκτενή βιβλιογραφία. Μάλιστα, βασικός στόχος είναι οι παρεχόμενοι αλγόριθμοι και υπολογιστικά εργαλεία να αποτελέσουν ένα οπλοστάσιο στα χέρια των βιοπληροφορικάριων για την επίτευξη της κατανόησης των κυτταρικών λειτουργιών και την χρησιμοποίηση αυτής της γνώσης για γονιδιακή θεραπεία διαφόρων πολύπλοκων πολυπαραγοντικών ασθενειών όπως ο καρκίνος. Τα σημαντικόταρα επιτεύγματα της παρούσας διατριβής μπορούν να συνοψισθούν στα ακόλουθα σημεία: • Παροχή ολοκληρωμένης υπολογιστικής διαδικασίας ανάλυσης δεδομένων πρωτεϊνικών αλληλεπιδράσεων • Σχεδιασμός και υλοποίηση ευφυών τεχνικών πρόβλεψης και βαθμολόγησης πρωτεϊνικών αλληλεπιδράσεων, που θα παρέχουν αποδοτικά και ερμηνεύσιμα μοντέλα πρόβλεψης. • Σχεδιασμός και υλοποίηση αποδοτικών αλγορίθμων μη επιβλεπόμενης μάθησης για την εξόρυξη πρωτεϊνικών συμπλόκων από δίκτυα πρωτεϊνικών αλληλλεπιδράσεων. • Δημιουργία μιας βάσης γνώσης που θα παρέχει στην επιστημονική κοινότητα όλα τα ευρήματα της ανάλυσης των δεδομένων πρωτεϊνικών αλληλεπιδράσεων για τον ανθρώπινο οργανισμό. / The present dissertation was conducted in the Pattern Recognition Laboratory, of the Department of Computer Engineering and Informatics at the University of Patras. It is a part of the wide research activity of the Pattern Recognition Laboratory in the domain of designing, implementing and applying Computational Intelligence technologies for the analysis of biological data. The present dissertation was co-financed by the research program Hrakleitos II. The proteomics domain is a quite new and fast evolving research domain. One of the great challenges in the domain of proteomics is the reconstruction of the complete protein-protein interaction network within the cells. The analysis of these networks is able to uncover the role of protein-protein interactions in diseases as well as their developmental procedure, as protein-protein interactions play very important roles in the basic cellular functions. However, this is very hard to be accomplished as protein-protein interactions and the whole proteome is differentiated among cells and it constantly changes through the biochemical cellular and environment interactions. An organism has radically different protein expression in different tissues, in different phases of his life and under varying environmental conditions. Two very important domains of research are created. First, the identification of the real protein-protein interactions within an organism which will compose its protein interaction network. Second, the analysis of the protein interaction network to extract knowledge (search for protein complexes, uncovering of proteins functionality e.tc.) In the present dissertation novel algorithmic Computational Intelligent techniques are presented for the prediction of protein-protein interactions, the prediction of a confidence score for each predicted protein-protein interaction, the prediction of protein complexes and the prediction of proteins functionality. In particular, in the task of predicting and scoring protein-protein interactions, a wide range of novel classification techniques was designed and developed. These techniques range from hybrid combinations of meta-heuristic methods and machine learning classifiers, to genetic programming methods and fuzzy systems. For the task of predicting protein complexes, two novel unsupervised methods were designed and developed which theoretically and experimentally surpassed the limitations of existing methodologies. For most of the designed techniques user friendly interfaces were developed to allow their utilizations by other researchers. Moreover, many of the implemented techniques were successfully applied to other research domaines such as the prediction of microRNAs and their targets and the forecastment of financial time series. The experimental procedure, initially focused on the well studied organism of Yeast (Saccharomyces cerevisiae) to validate the performance of the proposed algorithms and compare them with existing computational methodologies. Then, it focuses on the analysis of protein-protein interaction data from the Human organism. In specific, the best algorithmic techniques, from the ones proposed in the present dissertation, were applied to a human protein-protein interaction dataset. This resulted to the construction of a weighted protein-protein interaction network of high coverage, to the extraction of human protein complexes and to the functional characterization of Human proteins and complexes. The results of the analysis of Human protein-protein interaction data are available in the web knowledge base HINT-KB (http://hintkb.ceid.upatras.gr) which was implemented during this dissertation. In this knowledge base, structural, functional and sequential information from various sources were incorporated for every protein pair. Moreover, HINTKB provide access to the predicted and scored protein-protein interactions and to the predicted protein complexes and their functional characterization. The problems which occupied the present dissertation have very significant research interest as it is proved by the provided wide bibliography. The basic goal is the provided algorithms and tools to contribute in the ultimate goal of systems biology to understand the cellular mechanisms and contribute in the development of genomic therapy of complex diseases such as cancer. The most important achievements of the present dissertation are summarized in the next points: • Providing an integrated computational framework for the analysis of protein-protein interaction data. • Designing and implementing intelligent techniques for predicting and scoring protein-protein interactions in an accurate and interpretable manner. • Designing and implementing effective unsupervised algorithmic techniques for extracting protein complexes and predicting their functionality. • Creating a knowledge base which will provide to the scientific community all the findings of the analysis conducted on the Human protein-protein interaction data.

Στην παρούσα διατριβή παρουσιάζονται παράλληλοι αλγόριθμοι και εφαρμογές σε πολυπύρηνες μονάδες επεξεργασίας γραφικών. Πιο συγκεκριμένα, εξετάζονται οι μέθοδοι σχεδίασης ενός παράλληλου αλγορίθμου για την επίλυση τόσο απλών και κοινών προβλημάτων, όπως η ταξινόμηση, όσο και υπολογιστικά απαιτητικών προβλημάτων, έτσι ώστε να εκμεταλλευτούμε πλήρως την τεράστια υπολογιστική δύναμη που προσφέρουν οι σύγχρονες μονάδες επεξεργασίας γραφικών. Πρώτο πρόβλημα που εξετάστηκε είναι η ταξινόμηση, η οποία είναι ένα από τα πιο συνηθισμένα προβλήματα στην επιστήμη των υπολογιστών. Υπάρχει σαν εσωτερικό πρόβλημα σε πολλές εφαρμογές, επομένως πετυχαίνοντας πιο γρήγορη ταξινόμηση πετυχαίνουμε πιο καλή απόδοση γενικότερα. Στο Κεφάλαιο 3 περιγράφονται όλα τα βήματα σχεδιασμού για την εκτέλεση ενός αλγορίθμου ταξινόμησης για ακεραίους, της count sort, σε μια μονάδα επεξεργασίας γραφικών. Σημαντική επίδραση στην απόδοση είχε η αποφυγή του συγχρονισμού των νημάτων στο τελευταίο βήμα του αλγορίθμου. Στη συνέχεια παρουσιάζονται εφαρμογές παράλληλων αλγορίθμων σε υπολογιστικά απαιτητικά προβλήματα. Στο Κεφάλαιο 4, εξετάζεται το πρόβλημα χρονοπρογραμματισμού εξετάσεων Πανεπιστημίων, το οποίο είναι ένα πρόβλημα συνδυαστικής βελτιστοποίησης. Για την επίλυσή του χρησιμοποιείται ένας υβριδικός εξελικτικός αλγόριθμος, ο οποίος εκτελείται εξ' ολοκλήρου στην μονάδα επεξεργασίας γραφικών. Η τεράστια υπολογιστική δύναμη της GPU και ο παράλληλος προγραμματισμός δίνουν τη δυνατότητα χρήσης μεγάλων πληθυσμών έτσι ώστε να εξερευνήσουμε καλύτερα τον χώρο λύσεων και να πάρουμε καλύτερα ποιοτικά αποτελέσματα. Στο επόμενο κεφάλαιο γίνεται επίλυση του προβλήματος σχεδιασμού κίνησης για υποθαλάσσια οχήματα με βραχίονα. Εξετάζεται το πρόβλημα τόσο του ολικού σχεδιασμού όσο και του τοπικού. Στην πρώτη περίπτωση είναι σημαντική η καλή λύση και η ακρίβεια και ο παράλληλος αλγόριθμος που χρησιμοποιείται για την αναπαράσταση του περιβάλλοντος εργασίας σε μια Bump-επιφάνεια βοηθάει προς αυτή την κατεύθυνση. Στη δεύτερη περίπτωση, το πρόβλημα είναι πρόβλημα πραγματικού χρόνου και μας ενδιαφέρει η ταχύτητα εύρεσης της επόμενης θέσης του οχήματος. Ο παράλληλος προγραμματισμός και η GPU βοηθούν σημαντικά σε αυτό. Τελευταία εφαρμογή που εξετάστηκε είναι η μελέτη ενός συστήματος ημιφθοριωμένων αλκανίων με την μοριακή προσομοίωση Monte Carlo. Η παραλληλοποίηση ενός μέρους, του πιο χρονοβόρου, του αλγορίθμου έδωσε τη δυνατότητα εξέτασης ενός πολύ μεγαλύτερου συστήματος σε αποδεκτό χρόνο. Σε γενικές γραμμές, γίνεται φανερό ότι ο παράλληλος προγραμματισμός και οι σύγχρονες πολυπύρηνες αρχιτεκτονικές, όπως οι μονάδες επεξεργασίας γραφικών, δίνουν νέες δυνατότητες στην αντιμετώπιση καθημερινών προβλημάτων, προβλημάτων πραγματικού χρόνου και προβλημάτων συνδυαστικής βελτιστοποίησης. / In this thesis, parallel algorithms and applications in manycore graphics processing units are presented. More specifically, we examine methods of designing a parallel algorithm for solving both simple and common problems such as sorting, and computationally demanding problems, so as to fully exploit the enormous computing power of modern graphics processing units (GPUs). First problem considered is sorting, which is one of the most common problems in computer science. It exists as an internal problem in many applications. Therefore, sorting faster, results in better performance in general. Chapter 3 describes all design options for the implementation of a sorting algorithm for integers, count sort, on a graphics processing unit. The elimination of thread synchronization in the last step of the algorithm had a significant effect on the performance. Chapter 4 addresses the examination timetabling problem for Universities, which is a combinatorial optimization problem. A hybrid evolutionary algorithm, which runs entirely on GPU, was used to solve the problem. The tremendous computing power of GPU and parallel programming enable the use of large populations in order to explore better the solution space and get better quality results. In the next chapter, the problem of motion planning for underwater vehicle manipulator systems is examined. In the gross motion planning problem, it is important to achieve a good solution with high accuracy. The parallel algorithm used for the representation of the working environment in a Bump-surface is a step towards this direction. In the local motion planning problem, which is a real-time problem, the time needed to find the next configuration of the vehicle is crucial. Parallel programming and the GPU greatly assist in this online problem. Last application considered is the atomistic Monte Carlo simulation of semifluorinated alkanes. The parallelization of part of the algorithm, the most time-consuming, enabled the study of a much larger system in an acceptable execution time. In general, it becomes obvious that parallel programming and new novel manycore architectures, such as graphics processing units, give new capabilities for solving everyday problems, real time and combinatorial optimization problems.

Η διατριβή πραγματεύεται το αντικείμενο της Υπολογιστικής Νοημοσύνης στην Οικονομική και Χρηματοοικονομική επιστήμη. Στο πρώτο μέρος της διατριβής αναπτύσσονται μέθοδοι ομαδοποίησης και υπολογιστικής νοημοσύνης για τη μοντελοποίηση και πρόβλεψη χρονολογικών σειρών ημερησίων συναλλαγματικών ισοτιμιών. Η προτεινόμενη μεθοδολογία κατασκευάζει τοπικούς προσέγγιστές, με τη μορφή νευρωνικών δικτύων, για ομάδες προτύπων στο χώρο εισόδων που αναγνωρίζονται από μη-επιβλεπόμενους αλγόριθμους ομαδοποίησης. Στη συνέχεια κατασκευάζονται τεχνικοί κανόνες συναλλαγών απευθείας από τα δεδομένα με τη χρήση γενετικού προγραμματισμού. Η επίδοση των νέων κανόνων συγκρίνεται με αυτή των γενικευμένων κανόνων κινητού μέσου. Το δεύτερο μέρος της διατριβής πραγματεύεται την εφαρμογή εξελικτικών αλγορίθμων για τον υπολογισμό και την εκτίμηση του πλήθους σημείων ισορροπίας σε προβλήματα από τη θεωρία παιγνίων και τη νέα οικονομική γεωγραφία. Πιο συγκεκριμένα, αξιολογείται η ικανότητα των εξελικτικών αλγορίθμων να εντοπίσουν σημεία ισορροπίας κατά Nash σε πεπερασμένα στρατηγικά παίγνια και προτείνονται τεχνικές για τον εντοπισμό περισσοτέρων του ενός σημείων ισορροπίας. Τέλος εφαρμόζονται κριτήρια από τη θεωρία υπολογισμού σταθερών σημείων και τη θεωρία τοπολογικού βαθμού για τη διερεύνηση της ύπαρξης και της υπολογιστικής πολυπλοκότητας του υπολογισμού βραχυχρόνιων σημείων ισορροπίας σε μοντέλα νέας οικονομικής γεωγραφίας. / The thesis investigates Computational Intelligence methods in Economics and Finance. The first part of the thesis is devoted to computational intelligence methods and unsupervised clustering methods for modeling and forecasting daily exchange rate time series. A methodology is proposed that relies on local approximation, using artificial neural networks, for subregions of the input space that are identified through unsupervised clustering algorithms. Furthermore, we employ genetic programming to construct novel trading rules directly from the data. The performance of the novel rules is compared to that of generalised moving average rules. In the second part of the thesis we employ evolutionary algorithms to compute and to estimate the number of equilibria in finite strategic games and new economic geography models. In particular, we investigate the capability of evolutionary and swarm intelligence algorithms to compute Nash equilibria and propose an approach for the computation of more than one equilibria. Finally we employ criteria from the theory on computation of fixed points and topological degree theory to investigate the existence and the computational complexity of computing short run equilibria in new economic geography models.

No description available.

The complex co-evolutionary systems approach (CCeSA) provides a well-suited framework for analysing agricultural systems, serving as a bridge between biophysical and socioeconomic sciences, allowing for the explanation of phenomena, and for the use of metaphors for thinking and action. By studying agricultural systems as self-generated, hierarchical, complex co-evolutionary farming systems (CCeFSs), one can investigate the interconnections between the elements that constitute CCeFSs, along with the relationships between CCeFSs and other systems, as a fundamental step to understanding sustainability as an emergent property of the system. CCeFSs are defined as human activity systems emerging from the purposes, gestalt, mental models, history and weltanschauung of the farm manager, and from his dynamic co-evolution with the environment while managing the resources at his hand to achieve his own multiple, conflicting, dynamic, semi-structured and constrained purposes. A sustainable CCeFS is described as one that exhibits both enough fitness to achieve its multiple, dynamic, constrained, semi-structured, and often incommensurable and conflicting purposes while performing above threshold values for failure, and enough flexibility to dynamically co-evolve with its changing biophysical and socioeconomic environment for a given future period. Fitness and flexibility are essential features of sustainable CCeFSs because they describe the systems' dynamic capacity to explore and exploit its dynamic phase space while co-evolving with it. This implies that a sustainable CCeFS is conceived as a set of dynamic, co-evolutionary processes, contrasting with the standard view of sustainability as an equilibrium or steady state. Achieving sustainable CCeFSs is a semi-structured, constrained, multi-objective, and dynamic optimisation management problem with an intractable search phase space, that can be solved within the CCeSA with the help of a multi-objective co-evolutionary optimisation tool. Carnico-ICSPEA2, a Co-Evolutionary Navigator (CoEvoNav) used as a CCeSA's tool for harnessing the complexity of the CCeFS of interest and its environment towards sustainability, is introduced. The software was designed by its end-user - the farm manager and author of this thesis - as an aid for the analysis and optimisation of the "San Francisco" ranch, a beef cattle enterprise running on temperate pastures and fodder crops in the central plateau of Mexico. By combining a non-linear simulator and a multi-objective evolutionary algorithm with a deterministic and stochastic framework, the CoEvoNav imitates the co-evolutionary pattern of the CCeFS of interest. As such, the software was used by the farm manager to "navigate" through his CCeFS's co-evolutionary phase space towards achieving sustainability at farm level. The ultimate goal was to enhance the farm manager's decision-making process and co-evolutionary skills, through an increased understanding of his system, the co-evolutionary process between his mental models, the CCeFS, and the CoEvoNav, and the continuous discovery of new, improved sets of heuristics. An overview of the methodological, theoretical and philosophical framework of the thesis is introduced. Also, a survey of the Mexican economy, its agricultural sector, and a statistical review of the Mexican beef industry are presented. Concepts such as modern agriculture, the reductionist approach to agricultural research, models, the system's environment, sustainability, conventional and sustainable agriculture, complexity, evolution, simulators, and multi-objective optimization tools are extensively reviewed. Issues concerning the impossibility of predicting the long-term, detailed future behaviour of CCeFSs, along with the use of simulators as decision support tools in the quest for sustainable CCeFSs, are discussed. The rationale behind the simulator used for this study, along with that of the multi-objective evolutionary tools used as the makeup of Carnico-ICSPEA2, are explained. A description of the "San Francisco" ranch, its key on-farm sustainability indicators in the form of objective functions, constraints, and decision variables, and the semi-structured, multi-objective, dynamic, constrained management problem posed by the farm manager's planned introduction of a herd of bulls for fattening as a way to increase the fitness of his CCeFS via a better management of the system's feed surpluses and the acquisition of a new pick-up truck are described as a case study. The tested scenario and the experimental design for the simulations are presented as well. Results from using the CoEvoNav as the farm manager's extended phenotype to solve his multi-objective optimisation problem are described, along with the implications for the management and sustainability of the CCeFS. Finally, the approach and tools developed are evaluated, and the progress made in relation to methodological, theoretical, philosophical and conceptual notions is reviewed along with some future topics for research.

The Multi-Layer Perceptron (MLP) is one of the most widely applied and researched Artificial Neural Network model. MLP networks are normally applied to performing supervised learning tasks, which involve iterative training methods to adjust the connection weights within the network. This is commonly formulated as a multivariate non-linear optimization problem over a very high-dimensional space of possible weight configurations. Analogous to the field of mathematical optimization, training an MLP is often described as the search of an error surface for a weight vector which gives the smallest possible error value. Although this presents a useful notion of the training process, there are many problems associated with using the error surface to understand the behaviour of learning algorithms and the properties of MLP mappings themselves. Because of the high-dimensionality of the system, many existing methods of analysis are not well-suited to this problem. Visualizing and describing the error surface are also nontrivial and problematic. These problems are specific to complex systems such as neural networks, which contain large numbers of adjustable parameters, and the investigation of such systems in this way is largely a developing area of research. In this thesis, the concept of the error surface is explored using three related methods. Firstly, Principal Component Analysis (PCA) is proposed as a method for visualizing the learning trajectory followed by an algorithm on the error surface. It is found that PCA provides an effective method for performing such a visualization, as well as providing an indication of the significance of individual weights to the training process. Secondly, sampling methods are used to explore the error surface and to measure certain properties of the error surface, providing the necessary data for an intuitive description of the error surface. A number of practical MLP error surfaces are found to contain a high degree of ultrametric structure, in common with other known configuration spaces of complex systems. Thirdly, a class of global optimization algorithms is also developed, which is focused on the construction and evolution of a model of the error surface (or search spa ce) as an integral part of the optimization process. The relationships between this algorithm class, the Population-Based Incremental Learning algorithm, evolutionary algorithms and cooperative search are discussed. The work provides important practical techniques for exploration of the error surfaces of MLP networks. These techniques can be used to examine the dynamics of different training algorithms, the complexity of MLP mappings and an intuitive description of the nature of the error surface. The configuration spaces of other complex systems are also amenable to many of these techniques. Finally, the algorithmic framework provides a powerful paradigm for visualization of the optimization process and the development of parallel coupled optimization algorithms which apply knowledge of the error surface to solving the optimization problem.

Notwithstanding the recent successes of deep convolutional neural networks for classification tasks, they are sensitive to the selection of their hyperparameters, which impose an exponentially large search space on modern convolutional models. Traditional hyperparameter selection methods include manual, grid, or random search, but these require expert knowledge or are computationally burdensome. Divergently, Bayesian optimization and evolutionary inspired techniques have surfaced as viable alternatives to the hyperparameter problem. Thus, an alternative hybrid approach that combines the advantages of these techniques is proposed. Specifically, the search space is partitioned into discrete-architectural, and continuous and categorical hyperparameter subspaces, which are respectively traversed by a stochastic genetic search, followed by a genetic-Bayesian search. Simulations on a prominent image classification task reveal that the proposed method results in an overall classification accuracy improvement of 0.87% over unoptimized baselines, and a greater than 97% reduction in computational costs compared to a commonly employed brute force approach. / Electrical and Mining Engineering / M. Tech. (Electrical Engineering)

Submitted by Alison Vanceto (alison-vanceto@hotmail.com) on 2017-01-03T12:57:22Z No. of bitstreams: 1 DissMRS.pdf: 7075641 bytes, checksum: 8e6f815544b7f6f2ce4a1a5a47b25482 (MD5) / Approved for entry into archive by Marina Freitas (marinapf@ufscar.br) on 2017-01-16T16:33:09Z (GMT) No. of bitstreams: 1 DissMRS.pdf: 7075641 bytes, checksum: 8e6f815544b7f6f2ce4a1a5a47b25482 (MD5) / Approved for entry into archive by Marina Freitas (marinapf@ufscar.br) on 2017-01-16T16:33:38Z (GMT) No. of bitstreams: 1 DissMRS.pdf: 7075641 bytes, checksum: 8e6f815544b7f6f2ce4a1a5a47b25482 (MD5) / Made available in DSpace on 2017-01-16T16:33:48Z (GMT). No. of bitstreams: 1 DissMRS.pdf: 7075641 bytes, checksum: 8e6f815544b7f6f2ce4a1a5a47b25482 (MD5) Previous issue date: 2016-12-14 / Não recebi financiamento / In recent years, companies have used Artificial Intelligence (AI) techniques to facilitate the decisionmaking process in manufacturing systems. The use of these techniques allows increased performance of Flexible Manufacturing System (FMS). The automation of the process using computational resources allows a deeper analysis of the system conditions, which sometimes result in a better decision taking. In this sense, the Fuzzy Logic has been engaged to carry out this task, because it has the characteristic of dealing easily with inaccurate information and encoding knowledge specialist in Fuzzy rules. However, as soon as the system complexity increases, the task of generating a Fuzzy Rule Base (FRB) appropriate to the proposed system becomes increasingly difficult. To assist this process of generation of the FRB, several techniques can be used and among them stand out the search technique called Evolutionary Algorithm (EA). The EA is used, for example, for tuning the FRB of the FMS through the reduction of the optimization variables values as Makespan or Tardiness. In the case of variable called Makespan, the tuning occurs when the EA generates an FRB that reduces the makespan values of a FMS. However, the construction of the EA that effectively generates a tuning FRB is not trivial. It is required to be in the process, the construction of various EA with different selection methods and different mutation rates among other settings until an appropriate EA for a given situation appears. Therefore, in this study we aim to build an environment configuration and performance analysis of EAs in order to define the tuning FRB of the Fuzzy Control System of an FMS, i.e., it is intended to investigate how the EA ideal parameter scenario used for tuning the FRB of the said control system. In this study, the used EA was an extension of Genetic Algorithm (GA). For implementing the proposal, an evolutionary system for configuration and analysis of this variant of the GA was created. In this system, entitled "EvolSys - Evolutionary System" parameters of the system as Number of Input Variables of FRB, Number of Output Variables of FRB, Population Size, Mutation Rate and the EA Crossover Rate, among others are configured and then, one FRB is generated. Using this, there is an EA analysis of the possibility for choosing a FRB that will provide the reduction of makespan in FMS. Consequently, through this study, we may conclude that the use of EAs in collaboration with Fuzzy system may become an important tool for turning the system responsibility to the sequences of an FMS operation. Accordingly, the environment created meets the configuration step and analysis of EAs. / Nos últimos anos, empresas tem usado técnicas de Inteligência Artificial (AI) para auxiliar o processo de tomada de decisão em sistemas de manufatura. O uso dessas técnicas possibilita o aumento do desempenho dos Sistemas Flexíveis de Manufatura (FMS), uma vez que a automatização do processo com o uso de recursos computacionais permite uma análise mais profunda das condições do sistema o que, por vezes, resulta em uma melhor tomada de decisão. Neste sentido, a Lógica Fuzzy vem sendo usada para realizar essa tarefa, pois ela tem a característica de lidar facilmente com informações imprecisas, codificando o conhecimento do especialista nas chamadas Regras Fuzzy. Entretanto, à medida que a complexidade do sistema aumenta, a tarefa de gerar uma Base de Regras Fuzzy (FRB) adequada ao sistema proposto se torna cada vez mais difícil. Para auxiliar esse processo de geração da FRB, várias técnicas podem ser usadas e dentre elas destaca-se a técnica de busca denominada Algoritmo Evolutivo (EA). O EA pode ser usado, por exemplo, para a sintonia da Base de Regras Fuzzy do Sistema de Controle de um FMS por intermédio da redução de valores de variáveis de otimização como Makespan ou Tardiness. No caso da variável denominada Makespan, a sintonia ocorre quando o EA gera uma FRB que reduz os valores do makespan do FMS em questão. Entretanto, a construção do EA que efetivamente gera uma FRB sintonizada para um FMS não é trivial, pois é necessário que haja, nesse processo, a construção de vários tipos de EA com métodos de seleção diferentes, taxas de cruzamento e mutação diferentes dentre outras configurações, até que se encontre o EA adequado à uma dada situação. Sendo assim, no presente trabalho, o objetivo é a construção de um ambiente de configuração e análise de desempenho de EAs para sintonia da FRB do Sistema de Controle de um FMS, ou seja, pretende-se investigar qual o cenário de parâmetros ideal do EA usado na sintonia da FRB do referido sistema de controle. No presente trabalho, o EA usado foi uma extensão do Algoritmo Genético (GA). Para implementação da proposta, um Sistema Evolutivo para configuração e análise dessa variante do GA foi criado. Nesse sistema, intitulado “EvolSys - Evolutionary System”, parâmetros dos sistema como Número de Varáveis de Entrada da FRB, Número de Variáveis de Saída da FRB, Tamanho da População, Taxa de Mutação e Taxa de Cruzamento do EA, dentre outros são configurados e, por consequência, uma FRB é gerada. Com isso, há a possiblidade da análise do EA para a escolha de uma FRB que venha propiciar a redução do makespan em FMSs. Portanto, é possível concluir, a partir desse trabalho, que o uso de EAs em colaboração com os sistemas Fuzzy pode vir a se tornar uma importante ferramenta para sintonia da Base de Regras do sistema responsável pelo sequenciamento das operações de um FMS e, nesse sentido, o ambiente criado cumpre a etapa de configuração e análise do desempenho de EAs.

Made available in DSpace on 2014-12-17T14:52:49Z (GMT). No. of bitstreams: 1 WagnerO_DISSERT.pdf: 4964124 bytes, checksum: 34ed6ffd6dcd720ddf12631ffd06a3d6 (MD5) Previous issue date: 2010-03-17 / The telecommunications play a fundamental role in the contemporary society, having as one of its main roles to give people the possibility to connect them and integrate them into society in which they operate and, therewith, accelerate development through knowledge. But as new technologies are introduced on the market, increases the demand for new products and services that depend on the infrastructure offered, making the problems of planning of telecommunication networks become increasingly large and complex. Many of these problems, however, can be formulated as combinatorial optimization models, and the use of heuristic algorithms can help solve these issues in the planning phase. This paper proposes the development of a Parallel Evolutionary Algorithm to be applied to telecommunications problem known in the literature as SONET Ring Assignment Problem SRAP. This problem is the class NP-hard and arises during the physical planning of a telecommunication network and consists of determining the connections between locations (customers), satisfying a series of constrains of the lowest possible cost. Experimental results illustrate the effectiveness of the Evolutionary Algorithm parallel, over other methods, to obtain solutions that are either optimal or very close to it / As telecomunica??es desempenham um papel fundamental na sociedade contempor?nea, tendo como um de seus principais pap?is o de conceder ?s pessoas a possibilidade de conect?-las e integr?-las ? sociedade em que vivem e com isso acelerar o desenvolvimento por meio do conhecimento. Mas, ? medida que novas tecnologias s?o introduzidas no mercado, cresce tamb?m a demanda por novos produtos e servi?os que dependem da infraestrutura oferecida, tornando os problemas de planejamento de redes de telecomunica??es cada vez maiores e mais complexos. Muitos desses problemas, no entanto, podem ser formulados como modelos de Otimiza??o Combinat?ria, e o uso de algoritmos heur?sticos podem ajudar a solucionar essas quest?es da fase de planejamento. Este trabalho prop?e o desenvolvimento de um Algoritmo Evolutivo paralelo a ser aplicado ao problema de telecomunica??es conhecido na literatura por Problema de Atribui??o de Localidades a An?is em Redes SONET/SDH ou PALAS. Esse problema ? da classe NP-dif?cil e surge durante a etapa do planejamento f?sico da rede e consiste na determina??o das conex?es entre localidades (clientes), de modo a satisfazer uma s?rie de restri??es ao menor custo poss?vel. Os resultados dos experimentos ilustram a efici?ncia do Algoritmo Evolutivo paralelo, sobre outros m?todos, em obter solu??es ?timas ou muito pr?ximas do valor ?timo

Made available in DSpace on 2014-12-17T14:53:02Z (GMT). No. of bitstreams: 1 JoaoPLN_DISSERT.pdf: 2086259 bytes, checksum: 2a57554e118e00d9a44cdf572e29af7a (MD5) Previous issue date: 2010-12-23 / Coordena??o de Aperfei?oamento de Pessoal de N?vel Superior / This paper presents metaheuristic strategies based on the framework of evolutionary algorithms (Genetic and Memetic) with the addition of Technical Vocabulary Building for solving the Problem of Optimizing the Use of Multiple Mobile Units Recovery of Oil (MRO units). Because it is an NP-hard problem, a mathematical model is formulated for the problem, allowing the construction of test instances that are used to validate the evolutionary metaheuristics developed / O presente trabalho apresenta estrat?gias metaheur?sticas baseadas no framework dos Algoritmos Evolutivos (Gen?ticos e Mem?ticos) com a adi??o da t?cnica Vocabulary Building para a resolu??o do Problema de Otimiza??o do Emprego de Unidades M?veis de Pistoneio (UMPs). Por se tratar de um problema NP-?rduo, uma modelagem matem?tica ? formulada para o problema, permitindo a constru??o de inst?ncias testes que s?o utilizadas para validar as metaheur?sticas evolutivas desenvolvidas