Διδακτορικές διατριβές
Μόνιμο URI για αυτήν τη συλλογήhttps://pyxida.aueb.gr/handle/123456789/5
Περιήγηση
Πλοήγηση Διδακτορικές διατριβές ανά Τίτλο
A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
Α Β Γ Δ Ε Ζ Η Θ Ι Κ Λ Μ Ν Ξ Ο Π Ρ Σ Τ Υ Φ Χ Ψ Ω
Τώρα δείχνει 1 - 20 από 48
- Αποτελέσματα ανά σελίδα
- Επιλογές ταξινόμησης
Τεκμήριο Accelerating geometric queries for computer graphics: algorithms, techniques and applications(26-09-2024) Ευαγγέλου, Ιορδάνης; Evangelou, Iordanis; Athens University of Economics and Business, Department of Informatics; Fudos, Ioannis; Toumpis, Stavros; Markakis, Evangelos; Xylomenos, George; Dachsbacher, Carsten; Bittner, Jiri; Papaioannou, GeorgiosΣτο συνεχώς εξελισσόμενο τομέα των Γραφικών Υπολογιστών, η ζήτηση για ρεαλιστικά και σε πραγματικό χρόνο εικονικά περιβάλλοντα με αλληλεπίδραση σε ψηφιοποιημένα ή εκ γενετής ψηφιακά περιεχόμενα έχει εκθετικά αυξηθεί. Είτε πρόκειται για παιχνίδια, production rendering, computer-aided design reverse engineering, επεξεργασία και κατανόηση γεωμετρικής πληροφορίας ή διεργασίες προσομοίωσης, η ικανότητα γεωμετρικών αναζητήσεων οποιουδήποτε τύπου να εκτελούνται γρήγορα και με ακρίβεια είναι κρίσιμη. Η πραγματική μορφή ενός γεωμετρικού ερωτήματος ποικίλει ανάλογα με την εκάστοτε διεργασία, τον τομέα εφαρμογής, την αναπαράσταση της εισόδου και τη χρησιμοποιούμενη μεθοδολογία. Αυτά τα γεωμετρικά ερωτήματα μπορεί να περιλαμβάνουν ελέγχους τομής, όπως στην περίπτωση του ray tracing, χωρικά ερωτήματα, όπου χρειάζονται για την ανάκτηση πλησιέστερων γειτονικών δειγμάτων, γεωμετρική ανάκτηση, για την εξακρίβωση του τύπου των πολυγωνικών εισόδων ή ακόμα και κατανόηση μίας ολόκληρης εικονικής σκηνής για την ανάκτηση και ενσωμάτωση εναλλακτικών λύσεων, όπως στην περίπτωση βελτιστοποίησης και τοποθέτησης φωτιστικών πηγών. Καθώς οι εφαρμογές αυτών των αλγορίθμων και, συνεπώς, η πολυπλοκότητά τους αυξάνεται συνεχώς, τα βασικά αυτά γεωμετρικά ερωτήματα υστερούν όταν υιοθετούνται και ενσωματώνονται με απλοϊκό τρόπο σε πρακτικά σενάρια. Επομένως, αυτές οι μέθοδοι αντιμετωπίζουν περιορισμούς όσον αφορά την υπολογιστική αποδοτικότητα αυτών των αναζητήσεων. Αυτό είναι ιδιαίτερα έντονο σε σενάρια όπου πρέπει να ικανοποιηθεί ένας γιγάντιος αριθμός γεωμετρικών δεδομένων και ειδικότερα σε περιβάλλονται με διαδραστικό ή ακόμη και πραγματικό χρόνο απόδοσης. Τις περισσότερες φορές, πρέπει να εξετάσουμε και να κατανοήσουμε τους εσωτερικούς μηχανισμούς και τη θεωρία των αλγορίθμων που εκτελούν αυτά τα γεωμετρικά ερωτήματα. Αυτό είναι ιδιαίτερα χρήσιμο για την σχεδίαση κατάλληλων διαδικασιών προσαρμοσμένων στην εκάστοτε διεργασία, ώστε να μεγιστοποιηθεί η αποδοτικότητά τους, τόσο από άποψη ταχύτητας όσο και ποιότητας του αποτελέσματος. Αυτό έχει ως αποτέλεσμα, να υπάρχει ένας τεράστιος τομέας έρευνας που μελετά καινοτόμες προσεγγίσεις για την επιτάχυνση των γεωμετρικών αυτών αναζητήσεων, σε μία προσπάθεια να αντιμετωπίσει τις προκλήσεις που προκύπτουν.Ο κύριος στόχος αυτής της διατριβής ήταν να αναπτύξει καινοτόμες μεθόδους για την επιτάχυνση γεωμετρικών αναζητήσεων στο τομέα των Γραφικών Υπολογιστών. Η εργασία αυτή είχε μία ολιστική προσέγγιση στην διερεύνηση αλγοριθμικών βελτιστοποιήσεων που συντελούν στην ανάπτυξη προηγμένων δομών δεδομένων και αρχιτεκτονικών νευρωνικών δικτύων, ειδικά προσαρμοσμένων στην αποτελεσματική διαχείριση γεωμετρικών συλλογών. Η έρευνα αυτή όχι μόνο συμπεριέλαβε την υπολογιστική πολυπλοκότητα των αναζητήσεων αυτών, αλλά και την προσαρμοστικότητα των προτεινόμενων λύσεων σε διάφορες εφαρμογές και σενάρια, κυρίως στο πεδίο των Γραφικών Υπολογιστών αλλά και σε άλλους διασταυρούμενους τομείς. Τα αποτελέσματα αυτής της διατριβής έχουν τη δυνατότητα να επηρεάσουν τους τομείς και τις αντίστοιχες μεθοδολογίες που την υιοθετούν, αντιμετωπίζοντας τις σχετικές υπολογιστικές προκλήσεις και ανοίγοντας νέους ορίζοντες στην απεικόνιση πραγματικού χρόνου, την διαδραστική προσομοίωση και τις εμβυθιστικές τεχνολογίες εικονικής πραγματικότητας.Πιο συγκεκριμένα, οι συνεισφορές αυτής της διατριβής μπορούν να χωριστούν σε δύο κύριες κατευθύνσεις με στόχο την επιτάχυνση των γεωμετρικών αναζητήσεων: α) τεχνικές αναζήτησης πλησιέστερων δειγμάτων άμεσα επιταχυνόμενες από το, hardware στο πεδίο του ολικού φωτισμού και β) εφαρμογή τεχνικών βαθιάς μάθησης για τον ορισμό νέων δομών δεδομένων αλλά και γενικών μεθόδων γεωμετρικών αναζητήσεων.Οι μεθοδολογίες που ανήκουν στην πρώτη κατηγορία εξετάζουν την αλγοριθμική εφαρμογή ολικού φωτισμού σε πραγματικό χρόνο χρησιμοποιώντας εκτιμητές πυκνότητας των φωτονίων. Συγκεκριμένα, διερευνούμε σενάρια με δύσκολα φαινόμενα φωτιστικών ανακλάσεων, όπως αυτά των caustics που μπορούν να αντιμετωπιστούν κυρίως από την οικογένεια φωτιστικών μοντέλων του progressive photon mapping και απαιτούν την σκέδαση μεγάλου αριθμού ακτίνων τόσο από την κάμερα όσο και από τις φωτεινές πηγές. Τα φωτόνια που προέρχονται από τα φωτεινές πηγές, αποθηκεύονται στις επιφάνειες της γεωμετρίας ή σε κάποιο ογκομετρικό μέσο. Από εκεί θα πρέπει να συγκεντρωθούν μέσω αναζητήσεων που εκτελούνται κατά τις σκεδάσεις των ακτινών που προέρχονται από την κάμερα. Για την επίτευξη πραγματικού χρόνου απόκρισης ανά καρέ, η συγκέντρωση φωτονίων ανά αναζήτηση πρέπει να εκτελεστεί αποτελεσματικά παρά την χρονοβόρα λειτουργία που πρέπει να εκτελεστεί. Αυτό επιτυγχάνεται προσαρμόζοντας το tracing των ακτίνων σε μία screen space τεχνική και χρησιμοποιώντας το splatting ως μία μέθοδο άμεσα επιταχυνόμενη από το rasterisation. Τέλος, δεδομένου ότι η συγκέντρωση γειτόνων σε φραγμένο χώρο είναι μία εγγενής υποκατηγορία της αναζήτησης πλησιέστερων γειτόνων, προτείνουμε επιπλέον την αποτελεσματική γενίκευση αυτής της έννοιας σε οποιαδήποτε μορφή διεργασίας που εκμεταλλεύεται τα υπάρχοντα περιβάλλοντα ray tracing, και είναι άμεσα επιταχυνόμενα από το υπάρχον υλικό στις σημερινές κάρτες γραφικών. Έτσι, ενισχύουμε αποτελεσματικά τη φάση διεκπεραίωσης αυτών τον αναζητήσεων κατά πολλές τάξεις μεγέθους σε σύγκριση με τις υπάρχουσες παραδοσιακές στρατηγικές που συχνά υλοποιούνται.Όσον αφορά την δεύτερη κατηγορία των συνεισφορών μας, εστιάζουμε το ενδιαφέρον μας σε μια γενικότερη κατηγορία γεωμετρικών αναζητήσεων. Η πρώτη περιλαμβάνει την ακριβή και γρήγορη κατηγοριοποίηση γεωμετρικών σχημάτων χρησιμοποιώντας νευρωνικά δίκτυα. Συγκεκριμένα δείχνουμε ότι μια υβριδική μεθοδολογία επεξεργάζεται τον προσανατολισμό σε συνδυασμό με την voxelised γεωμετρική αναπαράσταση είναι ικανή να διεκπεραιώνει δύσκολα για κατηγοροιοποιήση στερεές γεωμετρίες που συναντούνται στο χώρο των κτιριακών κατασκευών. Δεύτερον, εξετάζουμε τη κλάση γεωμετρικών αναζητήσεων που σχετίζονται με την ανάλυση εικονικών σκηνών. Πιο συγκεκριμένα, μελετάμε την βελτιστοποίηση τοποθέτησης και ανάθεσης των φωτιστικών εντάσεων σε δρόμους αστικών οικισμών, που είναι εκ των πραγμάτων μια υπολογιστικά περίπλοκη διεργασία, ειδικά για μεγάλες εισόδους και αντιφατικούς περιορισμούς. Οι υπάρχουσες μεθοδολογίες που χρησιμοποιούνται στη πρόσφατη βιβλιογραφία συνήθως εισάγουν υποθέσεις ως προς την αναπαράσταση της εισόδου για να μετριάσουν τη πολυπλοκότητα αυτής της διεργασίας. Ωστόσο, σε αυτήν τη διατριβή, προσεγγίζουμε αυτό το πρόβλημα με μια ολιστική λύση που μπορεί να παράγει εφικτές και ποικίλες λύσεις σε πραγματικό χρόνο υιοθετώντας μία οικογένεια generative νευρωνικών δικτύων. Τέλος, προτείνουμε μια νέα και γενικευμένη προσέγγιση για την επίλυση αναδρομικών συναρτήσεων κόστους για την κατασκευή δομών επιτάχυνσης γεωμετρικών αναζητήσεων. Η συγκεκριμένη εργασία δείχνει προς μία νέα ερευνητική κατεύθυνση με σκοπό την κατασκευή δομών δεδομένων που καθοδηγούνται από αναδρομικές συναρτήσεις κόστους χρησιμοποιώντας μεθοδολογίες νευρωνικών δικτύων. Στόχος μας είναι να παρακάμψουμε την εξαντλητική αλλά και αδύνατη αναζήτηση του χώρου καταστάσεων δοσμένης της συνάρτησης κόστους, ώστε να δημιουργήσουμε υψηλής ποιότητας δομές δεδομένων για χωρικές αναζητήσεις.Τεκμήριο Algorithmic and game-theoretic aspects of computational social choice(31-03-2024) Papasotiropoulos, Georgios; Παπασωτηρόπουλος, Γεώργιος; Athens University of Economics and Business, Department of Informatics; Dimakis, Antonis; Pagourtzis, Aris; Amanatidis, Georgios; Filos-Ratsikas, Aris; Kavvadias, Dimitris; Sgouritsa, Alkmini; Markakis, EvangelosThe present dissertation aims to explore innovative decision-making approaches that complement traditional voting processes, examining them through an algorithmic, game-theoretic and axiomatic lens. The central objective is to identify voting procedures that can increase agents' desire to participate in collective governance and improve voters participation experience. Therefore, we focus on suggesting and analysing voting frameworks and election rules that reconcile the varying preferences of the electorate towards achieving socially desirable outcomes in various scenarios, aspiring to elevate both the quantity and quality of community involvement in democratic processes.Τεκμήριο Algorithmic problems in power management of computing systemsZois, Georgios; Athens University of Economics and Business, Department of Informatics; Universite Pierre et Marie Curie; Ecole doctorale Informatique, Telecommunications et Electronique, Laboratoire d' Informatique de Paris 6, Decision, Systemes Intelligents et Recherche Operationnelle; Bampis, Evripidis; Milis, IoannisThis thesis is focused on energy-efficient algorithms for job scheduling problems on speed-scalableprocessors, as well as on processors operating under a thermal and cooling mechanism, where, for a given budget of energy or a thermal threshold, the goal is to optimize a Quality of Service criterion. A part of our research concerns scheduling problems arising in large-data processing environments. In this context, we focus on the MapReduce paradigm and we consider problems of energy-efficient scheduling on multiple speed-scalable processors as well as classical scheduling on a set of unrelated processors.First, we study the minimization of the maximum lateness of a set of jobs on a single speed-scalable processor. We consider two variants of the problem: a budget variant, where we aim in minimizing maximum lateness for a given budget of energy and an aggregated variant, where we want to minimize a linear combination of maximum lateness and energy. We propose optimal algorithms for both variants in the non-preemptive case where jobs have common release dates. Our algorithms are based on a number of structural properties that can be obtained after applying the KKT (Karush-Kuhn-Tucker) conditions on a convex programming formulation of the problem. In the presence of arbitrary release dates, we prove that both variants become strongly NP-hard. Moreover, for the budget variant we show that it does not admit any O(1)-competitive deterministic algorithm, while for the aggregated variant we propose a 2-competitive online algorithm. Then, we study energy-aware MapReduce scheduling where the goal is to minimize the total weighted completion time of a set of MapReduce jobs under a given budget of energy. We first propose a convex programming relaxation of the problem, when the execution orderof jobs is known. We combine the solution of this relaxation with two natural list scheduling policies (First Come First Served and Highest Density First) and compare experimentally their effectiveness. Although their performance for random instances is fairly good, we prove that there are instances for which it is very far from the optimal. Next, we propose a linear programming approach which is based on an interval indexed LP-relaxation of the problem that incorporates a discretization of the possible speed values. Our algorithm transforms an optimal solution to this LP into a feasible solution for the problem by list scheduling in the order of tasks' _-points, where _ 2 (0; 1). We obtain a constant factorapproximation algorithm for the total weighted completion time of a set of MapReduce jobs using energy augmentation. In the context of classical MapReduce scheduling (where energy is not our concern) we also study the scheduling of a set of MapReduce jobs on unrelated processors with the goal of minimizing their total weighted completion time. We propose a 54-approximation algorithm which computes a feasible schedule by merging two individual schedules (of either Map or Reduce tasks) into a single schedule. Moreover, we consider the significant part of data shuffle in MapReduce applications and extend our model to capture the shuffle phase. We manage to keep the same ratio of 54 when the Shuffle tasks are scheduled on the same processors with the corresponding Reduce tasks, which becomes81 when the Shuffle and the Reduce tasks are scheduled on different processors. Finally, we focus on temperature-aware scheduling on a single processor that operates under a strict thermal threshold, where each job has its own heat contribution and the goal is to maximize the schedule's throughput. We consider the case of unit-length jobs with a common deadline and we revisit the offline CoolestFirst scheduling, i.e., the job with the smaller heat contribution is scheduled first. We study the approximability of Algorithm CoolestFirst and propose two different rounding schemes that yield lower bounds on its approximation factor. The first is based on a partition of the schedule according to theheat contributions of the jobs, while the second is based on a linear programming approach. The latter, which is actually more refined, yields a lower bound of at least 0.72.iΤεκμήριο Aspect Based Sentiment AnalysisPavlopoulos, Ioannis; Athens School of Economics and Business, Department of Informatics; Androutsopoulos, IonAspect Based Sentiment Analysis (ABSA) systems receive as input a set of texts (e.g.,product reviews or messages from social media) discussing a particular entity (e.g., anew model of a mobile phone). The systems attempt to detect the main (e.g., the mostfrequently discussed) aspects (features) of the entity (e.g., ‘battery’, ‘screen’) and toestimate the average sentiment of the texts per aspect (e.g., how positive or negativethe opinions are on average for each aspect). Although several ABSA systems have been proposed, mostly research prototypes, there is no established task decompositionfor ABSA, nore are there any established evaluation measures for the subtasks ABSA systems are required to perform. This thesis, proposes a new task decomposition for ABSA, which contains three main subtasks: aspect term extraction, aspect term aggregation, and aspect term polarity estimation. The first subtask detects single- and multi-word terms naming aspects of the entity being discussed (e.g., ‘battery’, ‘hard disc’), called aspect terms. The second subtask aggregates (clusters) similar aspect terms (e.g., ‘price’ and ‘cost’, but maybe also ‘design’ and ‘color’), depending on user preferences and other restrictions (e.g., the size of the screen where the results of the ABSA system will be shown). The third subtask estimates the average sentiment per aspect term or cluster of aspect terms. For each one of the above mentioned subtasks, benchmark datasets for different kinds of entities (e.g., laptops, restaurants) were constructed during the work of this thesis. New evaluation measures are introduced for each subtask, arguing that they are more appropriate than previous evaluation measures. For each subtask, the thesis also proposes new methods (or improvements over previous methods), showing experimentally on the constructed benchmark datasets that the new methods (or the improved versions) are better or at least comparable to state of the art ones.Τεκμήριο The contribution of experiential and material purchases to happiness: expanding the experience recommendation(2018) Kousi, Sofia; Baltas, George; Indounas, Konstantinos; Andronikidis, Andreas; Dimitriadis, Sergios; Siοmkos, George J.; Skarmeas, Dionysios; Kokkinaki, FloraThe experience recommendation posits that experiential purchases, such as concert and airplane tickets, contribute more to happiness than material purchases, such as clothes and cars. Closer examination of the extant research on the topic and the empirical explanations provided for this effect, reveals underlying asymmetries in the nature of these purchases, which contribute to the happiness advantage of experiential purchases. As such, the experience recommendation seems to only partially explain the relationship between consumption and happiness, thus limiting the value of any recommendation for maximizing happiness from spending. Τhe present research proposes and examines two asymmetries which give experiential purchases a happiness advantage. Specifically, the present study hypothesizes that experiential purchases are inherently more hedonic and self-expressive than material purchases. More importantly, it hypothesizes that once these asymmetries are corrected, material purchases elicit equal levels of happiness as experiential purchases. Through a qualitative study and five experiments, it is demonstrated that recalled experiential purchases tend to be more hedonic and self-expressive than material purchases. These characteristics give experiential purchases a happiness advantage, as they mediate the relationship between happiness and purchase type. The results of the present research contribute to the refinement of the experience recommendation, and indicate that under specific circumstances, material purchases can also make people happy. These findings refine the experience recommendation, deepen our understanding of the relationship between consumption and happiness, and add to a growing volume of research that reveals the complexity of this relationship. The data points to a necessary shift away from the question of whether experiential or material purchases make people happier, to asking what elements and characteristics of any purchase are likely to elicit happiness. The results suggest a prioritization of highly self-expressive and hedonic purchases, as these two characteristics positively influence the elicited happiness.Τεκμήριο Deep neural networks for biomedical question answering(01-05-2023) Παππάς, Δημήτριος; Pappas, Dimitrios; Athens University of Economics and Business, Department of Informatics; Papageorgiou, Haris; Koutsopoulos, Iordanis; Vassalos, Vasilios; Kotidis, Υannis; Konstas, Ioannis; Paliouras, Georgios; Androutsopoulos, IonΣε αυτήν τη διατριβή, επεκτείνουμε την έρευνα στον τομέα των ερωταποκρίσεων σε βιοϊατρικά κείμενα. Το πρώτο μέρος της διατριβής επικεντρώνεται στην ανάκτηση περιλήψεων επιστημονικής βιβλιογραφίας με βάση μια ερώτηση που υποβάλλεται από έναν βιοϊατρικό ειδικό εκφρασμένη σε φυσική γλώσσα. Προτείνουμε νέα μοντέλα βαθιάς μάθησης για την ανάκτηση βιοϊατρικών κειμένων και εξαγωγή αποσπασμάτων σχετικών με την ερώτηση. Προτείνονται επίσης μοντέλα που ανακτούν τα κείμενα και τα αποσπάσματα ταυτόχρονα, τα οποία βελτιώνουν ακόμη περισσότερο τα αποτελέσματα. Το καλύτερο μοντέλο βαθιάς μάθησης που δημιουργήθηκε έγινε δημόσια διαθέσιμο ως πρωτότυπο έρευνας κατά τη διάρκεια της πανδημίας του κορονοϊού για να βοηθήσει τους ερευνητές σε όλο τον κόσμο. Σε συνεργασία με βιοϊατρικούς ειδικούς, χρησιμοποιήσαμε ένα από τα μοντέλα βαθιάς μάθησης μας για την ανάκτηση εγγράφων και αναπτύξαμε ένα σύστημα αναγνώρισης λογοτεχνίας για συστηματικές ανασκοπήσεις.Ένα μείζον θέμα στις ερωταποκρίσεις σε βιοϊατρικά κείμενα είναι η έλλειψη επισημειωμένων δεδομένων από ανθρώπους, καθώς η επισημείωση απαιτεί ανθρώπινη ειδίκευση και χρόνο. Για αυτό το λόγο στο δεύτερο τμήμα της διατριβής, κατασκευάσαμε δύο νέα τεχνητά σύνολα δεδομένων για ερωταποκρίσεις σε βιοϊατρικά δεδομένα με τη μορφή πολλαπλής επιλογής (cloze-style QA) τα οποία και διατίθενται ελεύθερα. Ακολουθήσαμε μια μεθοδολογία που χρησιμοποιήθηκε στο παρελθόν σε ειδησεογραφικά άρθρα και βιβλία και κατασκευάζουμε εκατομμύρια τεχνητά παραδείγματα εκπαίδευσης που μπορούν να χρησιμοποιηθούν για την εκπαίδευση μοντέλων βαθιάς μάθησης που απαιτούν μεγάλο όγκο δεδομένων. Αξιολογώντας τις απαντήσεις των ανθρώπων στις κατασκευασμένες ερωτήσεις, δείχνουμε ότι οι άνθρωποι που είναι ειδικοί στον τομέα ξεπερνούν τους μη ειδικούς, το οποίο υποστηρίζει τον ισχυρισμό ότι η ανθρώπινη τεχνογνωσία είναι απαραίτητη για την απάντηση ερωτήσεων στον βιοϊατρικό τομέα. Αναπτύξαμε και εκπαιδεύσαμε νέα μοντέλα βαθιάς μάθησης για κατανόηση γραπτού λόγου χρησιμοποιώντας τα νέα μας σύνολα δεδομένων. Τα μοντέλα μας ξεπερνούν προηγούμενα προτεινόμενα μοντέλα βαθιάς μάθησης για ερωταποκρίσεις πολλαπλών επιλογών, καθώς και τέσσερα ισχυρά μοντέλα που δεν χρησιμοποιούν βαθιά μάθηση. Σε πειράματα που διεξήχθησαν σε ένα δείγμα του συνόλου δεδομένων, το καλύτερο μοντέλο ξεπέρασε όλους τους ανθρώπους που δεν ήταν ειδικοί και πέτυχε ανταγωνιστικά αποτελέσματα σε σύγκριση με τους ειδικούς βιοϊατρικής.Στο τρίτο μέρος της διατριβής, εκπαιδεύσαμε μοντέλα βαθιάς μάθησης για εξαγωγή απαντήσεων σε βιοϊατρικές ερωτήσεις ως τμήματα κειμένου (φράσεις) σε δύο καλά εδραιωμένα βιοϊατρικά σύνολα δεδομένων. Με δεδομένο ένα απόσπασμα κειμένου και μια ερώτηση, επιλέγεται ως απάντηση ένα τμήμα του αποσπάσματος.Εξετάζουμε έξι τεχνικές επαύξησης δεδομένων που εφαρμόζεται πριν από την εκπαίδευση των μοντέλων.Δείχνουμε ότι όλες οι τεχνικές επαύξησης δεδομένων βελτιώνουν την απόδοση, ακόμα και όταν χρησιμοποιούνται ως δεδομένα για περαιτέρω εκπαίδευση (fine-tuning) όταν χρησιμοποιούνται μεγάλα προεκπαιδευμένα γλωσσικά μοντέλα.Δείχνουμε επίσης ότι η χρήση ενός από τα τεχνητά σύνολα δεδομένων που δημιουργήθηκαν σε αυτή τη διατριβή λειτουργεί ως μια καλή τεχνική επαύξησης δεδομένων.Τεκμήριο Deep neural networks for information mining from legal texts(04/19/2021) Chalkidis, Ilias; Χαλκίδης, Ηλίας; Athens University of Economics and Business, Department of Informatics; Aletras, Nikolaos; Vazirgiannis, Michalis; Vassalos, Vasilios; Koubarakis, Manolis; Konstas, Ioannis; Paliouras, Georgios; Androutsopoulos, IonLegal text processing (Ashley, 2017) is a growing research area where Natural Language Processing (NLP) techniques are applied in the legal domain. There are several applications such as legal text segmentation (Mencia, 2009; Hasan et al., 2008), legal topic classification (Mencia and Fürnkranzand, 2007; Nallapati and Manning, 2008), legal judgment prediction and analysis (Wang et al., 2012; Aletras et al., 2016), legal information extraction (Kiyavitskaya et al., 2008; Dozier et al., 2010; Asooja et al., 2015), and legal question answering (Kim et al., 2015b, 2016b). These applications and relevant NLP techniques arise from three main sub-domains, i.e, legislation, court cases, and legal agreements (contracts). In all three sub-domains, documents are much longer than in most other modern NLP applications. They also have different characteristics concerning the use of language, the writing style, and their structuring, compared to non-legal text. Given the rapid growth of deep learning technologies (Goodfellow et al., 2016; Goldberg, 2017), the goal of this thesis is to explore and advance deep learning methods for legal tasks, such as contract element and obligation extraction, legal judgment prediction, legal topic classification, and information retrieval, that have already been discussed in the literature (but not in the context of deep learning) or that were first addressed during the work of this thesis. In this direction, we aim to answer two main research questions: First and foremost on the adaptability of neural methods that have been proposed for related NLP tasks in other domains and how they are affected by legal language, writing, and structure; and second on providing explanations of neural models’ decisions (predictions). Considering the first research question we find and highlight several cases, where either legal language affects a model’s performance or suitable modeling is needed to imitate the document structure. To this end, we pre-train and use in-domain word representations and neural language models, while we also propose new methods with state-of-the-art performance. With respect to model explainability, we initially experiment with saliency (attention) heat-maps and highlight their limitations as a means for the explanation of the model’s decisions, especially in the most challenging task of legal judgment prediction, where it is most important. To overcome these limitations we further study rationale extraction techniques as a prominent methodology towards model explainability.In lack of publicly available annotated datasets in order to experiment with deep learning methods, we curate and publish five datasets for various legal tasks (contract element extraction, legal topic classification, legal judgment prediction and rationale extraction, and legal information retrieval), while we also publish legal word embeddings and a legal pre-trained language model to assist legal text processing research and development. We consider our work, a first, fundamental, step among other recent efforts, towards improving legal natural language understanding using state-of-the-art deep learning techniques, which further promotes the adaptation of new technologies and sheds light on the emerging field of legal text processing.Τεκμήριο Design and analysis of auction mechanisms: algorithms and incentives(22-05-2023) Τσικιρίδης, Αρτέμ; Tsikiridis, Artem; Athens University of Economics and Business, Department of Informatics; Dimakis, Antonios; Karagiannis, Ioannis; Sgouritsa, Alkmini; Stamoulis, Georgios; Pagourtzis, Aris; Fotakis, Dimitris; Markakis, EvangelosΣε αυτή τη διατριβή, σχεδιάζουμε νέους αλγορίθμους για περιβάλλοντα συνδυαστικών δημοπρασιών ακολουθώντας μια διεπιστημονική προσέγγιση. Ταυτόχρονα, αναλύουμε την απόδοση υπαρχόντων πρωτοκόλλων δημοπρασιών και αναδεικνύουμε τις σχεδιαστικές αρχές εκείνες που επιτρέπουν εγγυήσεις απόδοσης.Στο πρώτο κομμάτι της διατριβής μελετάμε δύο υποδείγματα δημοπρασιών σημαντικών ως προς τις πρακτικές εφαρμογές τους: δημοπρασίες πυρήνα (core-selecting auctions) και δημοπρασίες πολλών αντιγράφων ενός αντικειμένου (multi-unit auctions). Αρχικά μελετούμε την έννοια του πυρήνα, όπως ορίστηκε από τους Ausubel και Milgrom. Μελετούμε το πολύτοπο που σχηματίζει ο πυρήνας σε μεγαλύτερο βάθος και αναδεικνύουμε μερικές νέες ιδιότητες. Χρησιμοποιώντας τις ιδιότητες αυτές, προτείνουμε έναν φιλαλήθη μηχανισμό που είναι ανταγωνιστικός ως προς τα MRCS έσοδα. Ο μηχανισμός αυτός είναι ο πρώτος ντετερμινιστικός, ανταγωνιστικός προς τον πυρήνα μηχανισμός για δυαδικά περιβάλλοντα δημοπρασιών μίας παραμέτρου στη βιβλιογραφία. Ακόμη, δίνουμε μια καταφατική απάντηση στην ερώτηση που είχε τεθεί στην βιβλιογραφία σχετικά με το αν υπάρχουν μη φθίνοντες (non-decreasing) MRCS μηχανισμοί. Στη συνέχεια, επικεντρωνόμαστε στις δημοπρασίες πολλών αντιγράφων ενός αντικειμένου (multi-unit auctions). Αναλύουμε δημοπρασίες διακριτής τιμής (discriminatory price), οι οποίες αποτελούν φυσική γενίκευση των δημοπρασιών πρώτης τιμής. Εξάγουμε νέα κάτω και άνω φράγματα ως προς το Τίμημα της Αναρχίας των μικτών σημείων ισορροπίας. Επιπλέον, παρουσιάζουμε έναν διαχωρισμό της κλάσης αυτής με την κλάση των Μπεϋζιανών σημείων ισορροπίας κατά Nash.Στο δεύτερο κομμάτι της διατριβής, μελετάμε δημοπρασίες προμηθειών (procurement auctions). Αρχικά, μελετάμε ένα πρόβλημα κάλυψης που προκύπτει σε γεωγραφικά μοντέλα αγορών πληθοπορισμού. Σχεδιάζουμε έναν φιλαλήθη μηχανισμό που πετυχαίνει έναν φραγμένο λόγο προσέγγισης σε σχέση με το βέλτιστο κόστος του δημοπράτη, βελτιώνοντας το καλύτερο γνωστό αποτέλεσμα της βιβλιογραφίας. Για την ίδια αντικειμενική συνάρτηση, σχεδιάζουμε έναν φιλαλήθες Πλήρως Πολυωνυμικού Χρόνου Σχήμα Προσέγγισης (FPTAS) για την περίπτωση εισόδων με σταθερό αριθμό εργασιών. Στη συνέχεια μελετάμε μια οικογένεια αντίστροφων δημοπρασιών στην οποία ο δημοπράτης έχει περιορισμένο προϋπολογισμό και οι πλειοδότες μπορούν να ανατεθούν να εκτελέσουν το καθήκον τους τμηματικά ή σε πολλά επίπεδα υπηρεσίας. Προτείνουμε δύο μηχανισμούς, έναν για κάθε περιβάλλον.Τεκμήριο Developing resilience and cyber-physical protection capabilities for critical aviation infrastructures(06/14/2021) Lykou, Georgia; Λύκου, Γεωργία; Athens University of Economics and Business, Department of Informatics; Apostolopoulos, Theodoros; Stamatiou, Υannis; Μαυρίδης, Ιωάννης; Μάγκος, Εμμανουήλ; Κοτζανικολάου, Παναγιώτης; Στεργιόπουλος, Γεώργιος; Gritzalis, DimitriosTransport sector is a critical infrastructure that greatly supports the smooth functioning of society's welfare and viability of economies worldwide. Disruptions to transportation systems can cause large economic impacts or even human losses, so they should be adequately protected from physical and cyber-physical threats.We focused our research on aviation sector, which is the safest transport mode, however the most interdepended one in terms of information and communication technologies applied. Cyber-attacks are increasing in quantity and persistence, so the consequences of a successful malicious cyber-attack to civil aviation operations could be severe nowadays. Aiming to enhance operational practices and develop robust cybersecurity governance in smart airports, we have presented a systematic and comprehensive analysis of unlawful attacks towards smart airports, by implementing cybersecurity best practices and resilience measures.Our research examined cyber security challenges and interoperability in Air Traffic Management systems and proposed an extended threat model for analysing possible targets and risks involved. We analysed cyber resilience aspects in the aviation context and the need for holistic strategy of defence, prevention, and response. Furthermore, as the fastest growing segment of aviation, Unmanned Aerial Systems(UAS) continue to increase in technical complexity, and capabilities. However, UAS pose significant challenges in terms of safety, security, and privacy. An increasing phenomenon, nowadays, is drone-related incidents near airport facilities, which are expected to proliferate in frequency and severity, as drones become larger and more powerful. Critical infrastructures need to be protected from such aerial attacks, through effective counteracting technologies, risk management, and resilience plans. In this dissertation, we have explored how counter drone technologies can prevent, detect, identify, and mitigate rogue drones. We have analysed realistic attack scenarios of malicious drones’ attacks and proposed an effective C-UAS protection plan for each case. We have also discussed the applicability limitations of C-UAS in the aviation context and proposed a resilience action plan for airports stakeholders for defending airborne threats from misused drones.The integration of our research in the aviation sector, focused on air transport networks and introduced a risk-based method to analyse interdependencies and congestions in the aviation network. The proposed methodology and software tool can assess delay incidents in airports, produce weighted risk dependency graphs, presenting how a delay that occurred in one airport may affect the operational efficiency of other interconnected airports. The tool can also detect the most critical airports and congested connections, while it can indicate the n-order dependency chains, which should be avoided by airline flight planners, to reduce delay impacts in the aviation network.Τεκμήριο Effective methods to tackle the equivalent mutant problem when testing software with mutationKintis, Marinos; Athens University of Economics and Business, Department of Informatics; Malevris, NikosMutation Testing is undoubtedly one of the most effective software testing techniques that has been applied to different software artefacts at different testing levels. Apart from mutation’s versatility, its most important characteristic is its ability to detect real faults. Unfortunately, mutation’s adoption in practice is inhibited, primarily due to the manual effort involved in its application. This effort is attributed to the Equivalent Mutant Problem. The Equivalent Mutant Problem is a well-known impediment to mutation’s practical adoption that affects all phases of its application. To exacerbate the situation, the Equivalent Mutant Problem has been shown to be undecidable in its general form. Thus, no complete, automated solution exists. Although previous research has attempted to address this problem, its circumvention remains largely an open issue. This thesis argues that effective techniques that considerably ameliorate the problem’s adverse effects can be devised. To this end, the thesis introduces and empirically evaluates several such approaches that are based on Mutant Classification, Static Analysis and Code Similarity. First, the thesis proposes a novel mutant classification technique, named Isolating Equivalent Mutants (I-EQM) classifier, whose salient feature is the utilization of second order mutants to automatically isolate first order equivalent ones. The empirical evaluation of the approach, based on real-world test subjects, suggests that I-EQM outperforms the existing techniques and results in a more effective testing process. Second, the thesis formally defines nine data flow patterns that can automatically detect equivalent and partially equivalent mutants. Their empirical evaluation corroborates this statement, providing evidence of their existence in real-world software and their equivalent mutant detection capabilities. Third, MEDIC (Mutants’ Equivalence Discovery), an automated framework that implements the aforementioned patterns and manages to detect equivalent and partially equivalent mutants in different programming languages, is introduced. The experimental investigation of the tool, based on a large set of manually analyzed mutants, reveals that MEDIC can detect efficiently more than half of the considered equivalent mutants and provides evidence of automated stubborn mutant detection. Finally, the thesis proposes the concept of mirrored mutants, that is mutants affecting similar code fragments and, more precisely, analogous code locations within these fragments. It is postulated that mirrored mutants exhibit analogous behavior of this concept supports this statement and suggests that the number of the equivalent mirrored mutants that have to be manually analyzed can be reduced approximately by halfΤεκμήριο Efficient texture representation and sampling algorithms for real-time rendering(Athens University of Economics and Business, 04-2013) Mavridis, Pavlos; Athnes University of Economics and Business, Department of Informatics; Papaioannou, GeorgiosDoctoral Thesis - Athnes University of Economics and Business.Τεκμήριο Energy efficient and adaptive service advertisement, discovery and provision for Mobile Ad Hoc NetworksVerveridis, Christopher N.; Athens University of Economics and Business, Department of Informatics; Polyzos, GeorgeMobile Ad Hoc Networks (MANETs) are networks of mobile computing nodes (e.g. portable computers, PDAs etc.) equipped with wireless interfaces and communicating with each other without relying on any infrastructure. In these networks each mobile node may act as a client, a server and a router. MANETs have emerged to fulfill the need for communication of mobile users in locations where deploying a network infrastructure is impossible, or too expensive, or simply is not available at that time. Characteristic scenarios for MANETs are disaster relief operations, battlefields and locations where infrastructure-based WLAN coverage (also called hotspots) is not provided and wireless WANs (e.g. GPRS/UMTS) are too expensive to use or too slow. Most of the research on MANETs has focused on issues dealing with the connectivity between mobile nodes in order to cope with the dynamism of such networks and the arising problems thereof. This dynamism is due to the mobility of nodes, the wireless channel’s adverse and fast changing conditions and the energy limitations of mobile nodes, all of which lead to frequent disconnections and/or node failures. These research efforts have led to the creation of a sound technical basis for dealing with the aforementioned problems regarding node connectivity in MANETs (mainly through routing protocols, link layer protocols etc.). However, solving the problems of connectivity alone is not sufficient for the adoption of MANETs. Since their basic role is to allow mobile users to exchange data and use each other’s services, there is also a need for architectures, mechanisms and protocols for Service Discovery and Provision. Service Discovery is defined in general as the process allowing networked entities to: i) advertise their services, ii) query about services provided by other entities, iii) select the most appropriate services and iv) invoke the services. Service Discovery has been mainly addressed in the context of wired networks in the past. In that context, clients and especially servers are typically powerful and resource-rich machines connected to the wired network. Service discovery and advertisement for wired networks generally follows a centralized or semi-centralized architecture assuming that well-known and “always on” service registries (or directories) exist for matching service requests to available services. Most of these architectures also rely on inefficient flooding techniques for service discovery and advertisement since resource scarcity is not a key issue. All the characteristics mentioned above, render those architectures inappropriate for service discovery in MANETs. In this context, the requirements are radically different than those of wired networks. In MANETs both clients and servers are more lightweight devices with limited resources. The assumption of (possibly dedicated and) always-available nodes serving as service directories is no longer valid. Also, frequent disconnections are a major issue affecting service availability. Disconnections may happen due to node mobility, due to depletion of the energy resources of the nodes, or even due to switching-off of some nodes. It is rather straightforward that in such environments it is of utmost importance to have energy efficient service discovery mechanisms. In this thesis we develop two energy efficient service discovery protocols integrated into the routing process in order to avoid redundant network messaging. Both developed protocols are distributed in nature and employ both proactive and reactive information dissemination techniques. The performance of the developed protocols is thoroughly investigated through extensive simulations in terms of energy consumption, service discoverability and achievable service availability under a wide range of settings (e.g. mobility, network density, channel characteristics, service replication, traffic patterns). The first (basic) protocol, called Extended Zone Routing Protocol (E-ZRP) is used in our study for comparing cross-layer with application-layer based service discoveryprotocols. Although integrating service and route discovery has been already proposed in the past, previous work regarding the energy gains of implementing service discovery at the routing layer as compared to implementing it at the application layer (as a separate process) is problematic due to the following reasons: i) most approaches either deal with energy only implicitly (by measuring the overhead in number of packets) or employ non-realistic energy consumption modeling, ii) the majority of proposed approaches involve performance analysis comparisons of the developed route and service discovery protocols against unrealistic application layer based service discovery protocols which are based on global flooding and hence are not suitable to MANETs, and iii) none of those approaches takes into consideration the impact of protocol message sizes on protocol performance. In our work we have performed fullstack simulations employing a realistic energy consumption model, accounting for the actual energy consumption of the nodes, in order to reveal the real gains from integrating the service discovery process with the routing process. The integrated service discovery protocol, named E-ZRP proves to be much more efficient compared to a similar but application layer based service discovery protocol. More importantly, we developed an even more sophisticated service discovery protocol, called Adaptive SerVicE and Route Discovery ProTocol for MANETs (AVERT). AVERT demonstrates superior efficiency than the best alternatives, mainly due to its built-in capability of adapting to a volatile environment such as a MANET. Using a monitoring mechanism for traffic seen locally, a node may adapt the operation of AVERT to best match the fluctuating traffic and mobility conditions typically found in MANETs. In parallel to the aforementioned research effort, we developed a novel mechanism for service providers to select to serve those clients from a MANET that can maximize their profits. Optimized service provisioning is a challenging problem in dynamic environments such as MANETs. We consider the nodes in MANETs to be independent, rational agents trying to maximize their profits through service provision. We model this problem as a Generalized Assignment Problem (GAP). We adopt a pay-as-you-go model, where clients pay for the service as long as they are receiving the service, since a pay-in-advance model would be unfair especially in MANETs where connection loss is very probable. We introduce into the proposed profit maximization algorithm expected payoffs based on estimates of server-to-client connectivity. Those estimations can be used for computing the actual payoff that will be received from any client that is selected by the service provider. We experimentally study cases with non-cooperative and cooperative servers and investigate the gain of the estimate based profit maximization algorithm versus a classic profit maximization algorithm, which does not take into account the network’s dynamics that affect server-to-client connectivity. We especially study the duration of connectivity (irrespectively of path changes) between two nodes of a MANET. Previous work, both analytical and experimental, has focused only on estimating the duration of a single path between a client and a server without considering changes to a path’s length caused by node mobility. The connectivity between two nodes irrespectively of changes in paths or the path length has not been investigated. In this dissertation we derive an approximation of the connectivity duration, taking into account the network’s density (number of nodes, terrain size, wireless transmission range), the node speed and the initial distance (in number of hops) between two nodes (one representing a client and the other a server). Simulations show that our approach achieves up to three-fold improved server profits compared to the classical one and is especially suited for MANETs with high-mobility and/or low density, which verifies that the proposed model accurately captures the effects of server to client connectivity on the overall performance. Summing up, in this thesis we propose energy efficient highly adaptive service discovery protocols integrated with hybrid routing protocols. We implemented and experimentally evaluated those protocols to show their energy and service discoverability/availability gains under different MANET conditions and against similar application layer-based as well as routing layer-based protocols. We also developed and evaluated an innovative mechanism for allowing mobile service providers to maximize their profits when offering their service in MANETs, by estimating their connectivity to candidate clients.Τεκμήριο Equilibrium and Optimal Control Policies for Queueing Models with Variable Service CapacityDimitrakopoulos, Yiannis; Athnes University of Economics and Business, Department of Informatics; Mageirou, Evangelos; Burnetas, ApostolosIn this thesis, we develop and analyze stochastic queueing models with which we study the effect of a varying service rate policy on customer arrival rate, considering two different ways for controlling customer entrance. In the first problem, we assume that customers respond strategically to excessive delays by taking into account the trade-off between service benefit and expected delay in the presence of a service control dynamic policy employed by the system administrator. In this case, arrival rate control is implicit, since it is imposed in equilibrium in the game between potential incoming customers. The effect of the service rate policy is that the possible equilibria are generally more than one. In the second problem, we assume that the system administrator controls both customer arrival rate and the service rate employed with respect to the system congestion in a stochastic dynamic framework. More specifically, in the first part of this dissertation, we consider the problem of customer equilibrium behavior on joining a single server Markovian queueing system with varying service rate between a low and a high value, according to a threshold based dynamic policy. We assume that arriving customers decide whether they will join a queue and receive a fixed reward or balk, i.e., decline to enter the queue, and compromise with an alternative payment. When customers make this decision they don't have any information on the system congestion upon arrival, but they know the implemented service policy. Since a tagged customer's expected delay is not monotone increasing in the arrival rate, the equilibrium strategy is not unique, and we show that the number of such strategies can vary from zero to three. Moreover, regarding the problem of maximizing the overall customer expected net benefit, we show numerically that, the optimal arrival rate that maximizes the customer's overall welfare lies between the extreme equilibrium arrival rates. Since the equilibrium arrival rates and the social optimal strategies do not coincide, we analyze this inefficiency considering the Price of Anarchy. We show that as the value of balking decreases, the price of anarchy becomes unbounded since customers are indifferent between joining and balking under any equilibrium strategy. The latter problem of social optimization can be viewed as one way of centrally controlling both admission and service rate. A more realistic setting to examine the impact of the service rate switch to a higher value on arrival rate is to assume that the system manager observes the queue length and dynamically sets both the arrival and the service rate. In the second part of this dissertation, we consider the effect of varying the service rate between a finite set of increasing values in an M/M/1 queue with admission control. Here, we formulate a Markov Decision Process model to analyze the joint admission and service control, considering both discounted and average expected profit maximization with the service payment being made upon admission or at the end of service. For each case, we derive a thresholdbased optimal policy for both controls. We define the value of service rate exibility as the difference in optimal profit between the joint control model and an admission control model with service rate set at the lowest value. Under a simple condition between the service payment and the relative cost of service rate increase, we show that the value of service rate exibility is positive, and, increasing in the system congestion. Moreover, the optimal admission threshold increases, so, more customers are accepted due to service rate exibility. Finally, sensitivity analysis with respect to the service payment for the different cases is pursued numerically.Τεκμήριο Fast and efficient predictions in Big Data systems(13-12-2023) Μαρούλης, Ευστάθιος; Maroulis, Stathis; Athens University of Economics and Business, Department of Informatics; Xylomenos, George; Gunopulos, Dimitrios; Voulgaris, Spyridon; Kotidis, Υannis; Stamoulis, Georgios; Hadjiefthymiades, Stathes; Kalogeraki, VanaΤην τελευταία δεκαετία, η έκρηξη στην παραγωγή δεδομένων από πηγές όπως τα social media, τα κινητά τηλέφωνα και τα δίκτυα αισθητήρων έχει ωθήσει την ανάπτυξη εφαρμογών που απαιτούν πολλά δεδομένα, συμπεριλαμβανομένων των συστημάτων συστάσεων και της πρόβλεψης κυκλοφοριακών μοτίβων. Αυτές οι εφαρμογές, που χειρίζονται δεδομένα σε κλίμακα GB ή TB, απαιτούν σημαντικούς υπολογιστικούς πόρους.Τα κατανεμημένα συστήματα επεξεργασίας μεγάλων δεδομένων όπως τα Apache Spark και Apache Flink είναι κρίσιμα στην επεξεργασία αυτών των τεράστιων όγκων δεδομένων. Ωστόσο, η βελτιστοποίηση αυτών των συστημάτων για αποτελεσματική λειτουργία των εφαρμογών και η ελαχιστοποίηση της κατανάλωσης ενέργειάς τους παρουσιάζει σημαντικές προκλήσεις. Σε αυτή την διατριβή αναπτύχθηκαν πρακτικές μεθοδολογίες για την αντιμετώπιση αυτών των προκλήσεων, εστιάζοντας στον αποδοτικό χρονοπρογραμματισμό (scheduling) εργασιών, στην ανάπτυξη μοντέλων πρόβλεψης υψηλής ακρίβειας, σε συστήματα ανίχνευσης συμβάντων και στην ενεργειακά αποδοτική χρήση πόρων.Η έρευνα αυτή περιλαμβάνει την εξέταση του τομέα του χρονοπρογραμματισμού πραγματικού χρόνου εργασιών σε κατανεμημένα συστήματα επεξεργασίας μεγάλων δεδομένων. Επικεντρώνεται επίσης στη δημιουργία ακριβών μοντέλων για την πρόβλεψη χρόνου εκτέλεσης εφαρμογών, λαμβάνοντας υπόψη τις πολυπλοκότητες όπως η ετερογένεια του περιβάλλοντος εκτέλεσης, οι παραμορφωμένες ενότητες δεδομένων και οι απαιτήσεις των εφαρμογών σε πραγματικό χρόνο, όπως ο χρόνος εκτέλεσης της εφαρμογής. Η προσέγγιση που ακολουθείται περιλαμβάνει την ανάπτυξη καινοτόμων αλγορίθμων χρονοπρογραμματισμού και ενός πρωτοποριακού συστήματος ανάλυσης (profiling). Επιπλέον, αντιμετωπίζει την πρόκληση του χρονοπρογραμματισμού εργασιών με έμφαση στην επίτευξη μετρικών απόδοσης ενώ ταυτόχρονα διαχειρίζονται συντηρητικά οι πόροι και η (ενεργειακή) κατανάλωση του συστήματος. Ένα άλλο σημαντικό επίτευγμα είναι η δημιουργία ενός συστήματος για την πρόβλεψη των μοτίβων κυκλοφορίας στην πόλη του Δουβλίνου, χρησιμοποιώντας τεχνικές Παραγοντοποίησης Πινάκων (Matrix Factorization) και εκμεταλλευόμενοι δεδομένα από αισθητήρες της πόλης. Τέλος, παρουσιάζεται ένα σύστημα που σχεδιάστηκε για την ταχεία και ακριβή παροχή συστάσεων και προβλέψεων στους χρήστες, εκμεταλλευόμενο δεδομένα από το Foursquare και χρησιμοποιώντας τεχνικές Παραγοντοποίησης Τανυστών (Tensor Factorization) για να προτείνει Σημεία Ενδιαφέροντος που προσαρμόζονται στις προτιμήσεις των χρηστών.Οι εμπειρικές αξιολογήσεις σε πραγματικά σύνολα δεδομένων αποδεικνύουν την αποτελεσματικότητα αυτών των μεθοδολογιών στη βελτίωση του χρονοπρογραμματισμού εργασιών, της πρόβλεψης της κυκλοφορίας και των συστημάτων συστάσεων, συμβάλλοντας σημαντικά στην προώθηση των δυνατοτήτων των διανεμημένων συστημάτων επεξεργασίας όπως το Apache Spark και το Apache Flink.Τεκμήριο Federated, continual and meta-learning algorithms for multi-task settings(13-12-2024) Νικολουτσόπουλος, Σωτήριος; Nikoloutsopoulos, Sotirios; Athens University of Economics and Business, Department of Informatics; Androutsopoulos, Ion; Koutsopoulos, Iordanis; Markakis, Evangelos; Blekas, Konstantinos; Stafylakis, Themos; Pavlopoulos, Ioannis; Toumpis, StavrosΗ ερευνητική εργασία η οποία περιλαμβάνεται στη διατριβή αυτή αφορά τρεις νέους αλγορίθμους Μηχανικής Μάθησης, ο καθένας εκ των οποίων αντιμετωπίζει προβλήματα πολλαπλών εργασιών (multiple tasks) σε ένα συγκεκριμένο πλαίσιο. Αυτού του είδους η Μάθηση πολλαπλών εργασιών (Multi-Task Learning) αφορά των χειρισμό πολλαπλών εργασιών είναι ταυτόχρονα είτε σειριακά, με κάθε εργασία να ορίζεται βάσει μιας συνάρτησης στόχου (objective function). Το πλήθος των εργασιών μπορεί να είναι πεπερασμένο ή άπειρο, και οι εργασίες μπορεί να είναι κατανεμημένες σε πολλούς πελάτες ή να βρίσκονται συγκεντρωμένες σε έναν. Οι αλγόριθμοι που προτείνει η ερευνητική εργασία αντιμετωπίσουν την μάθηση πολλαπλών εργασιών στα πλαίσια της προσωποποιημένης ομόσπονδης μάθησης, (Personalized Federated Learning), της Συνεχούς Μάθησης (Continual Learning) και της Μέτα-Μάθησης (Meta-Learning). Ο καθένας εξ αυτών είναι βελτιστοποιημένος για το συγκεκριμένο περιβάλλον στο οποίο λειτουργεί.Στο Κεφάλαιο 2 η διατριβή εξετάζει την Μάθηση Πολλαπλών Εργασιών στο πλαίσιο της προσωποποιημένης Ομόσπονδης Μάθησης και, συγκεκριμένα, εφαρμόζεται προκειμένου να κατασκευαστεί μια από κοινού αναπαράσταση (representation) στο εύρος ενός πεπερασμένου συνόλου εργασιών κατανεμημένων σε πολλαπλούς πελάτες. Αυτή η από κοινού αναπαράσταση για όλους τους πελάτες επιτρέπει καλύτερη γενίκευση και επιτρέπει σε κάθε ένα από τα μοντέλα να εκπαιδευτούν στην εργασία τους πιο γρήγορα, ιδιαιτέρως όταν το μέγεθος του συνόλου δεδομένων (dataset) του κάθε πελάτη είναι πεπερασμένο. Η διατριβή προτείνει έναν αλγόριθμο που βασίζεται σε μια ακριβή, κατανεμημένη, τεχνική στοχαστικής κατάβασης κλίσης (exact, distributed, Stochastic Gradient Descent) η οποία διαθέτει θεωρητικές εγγυήσεις σύγκλισης. Ο αλγόριθμος μπορεί να χειριστεί ένα πεπερασμένο πλήθος εργασιών κατανεμημένων σε πολλαπλούς πελάτες, μέσω της από κοινού βελτιστοποίησης δύο συνόλων βαρών, εκ των οποίων το ένα σύνολο περιλαμβάνει κοινά βάρη και το άλλο προσωποποιημένα βάρη. Η μέθοδος παρουσιάζει υψηλή επίδοση σε προβλήματα κατηγοριοποίησης σε σύνολα δεδομένων πολλαπλών κλάσεων (multi-class classification data sets)Στο Κεφάλαιο 3 η διατριβή εξετάζει την Μάθηση Πολλαπλών Εργασιών στο πλαίσιο της Συνεχούς Μάθησης (Continual Learning). Επομένως, σε αντίθεση με την συνήθη προσέγγιση κατά την οποία η μάθηση επιτελείται με πρόσβαση σε όλα τα δεδομένων εργασιών, ο πράκτορας πρέπει να προσαρμόζεται στις εργασίες ακολουθιακά, καθώς αυτές καταφθάνουν από μια ροή, ενώ ταυτόχρονα διατηρεί διαρκώς καλή επίδοση σε όλες τις εργασίες. Η διατριβή εστιάζει σε προβλήματα κατηγοριοποίησης και Συνεχή Μάθηση σταδιακά προστιθέμενων εργασιών (task-incremental), όπου νέες κλάσεις εισάγονται διαδοχικά, απαιτώντας από τον πράκτορα να προσαρμόζεται σε νέες κλάσεις καθώς περνά ο χρόνος. Σε αντίθεση με την απλή Μάθηση Πολλαπλών Εργασιών, στο πλαίσιο της Συνεχούς Μάθησης ο πράκτορας δεν έχει ταυτόχρονη πρόσβαση σε όλα τα δεδομένα και δεν μπορεί να επεξεργαστεί εκ νέου παλιά δεδομένα εκτός αν αυτά έχουν επί τούτου αποθηκευτεί στη μνήμη του. Αυτός ο περιορισμός, σε συνδυασμό με τις ανισορροπίες στην εμφάνιση των κλάσεων που είναι εγγενείς στη ροή των εισερχομένων δεδομένων, δημιουργεί δυσκολίες στο να επιτευχθεί δικαιοσύνη μεταξύ κλάσεων σε ό,τι αφορά την ακρίβεια (accuracy) καθώς ο πράκτορας πρέπει να μάθει να αντιμετωπίζει τις διαφορετικές κλάσεις δίκαια. Προκειμένου να αντιμετωπισθεί αυτή τη δυσκολία, η διατριβή εισάγει μια μέθοδο σύγχρονης (online) Συνεχούς Μάθησης που βασίζεται σε επαναληπτική μάθηση (online replay-based Continual Learning) που ελαχιστοποιεί την απώλεια Kullback-Leibler (KL) προκειμένου να επιτύχει δικαιοσύνη σε προβλήματα με μεγάλη ανισορροπία στα σχετικά πλήθη των παραδειγμάτων σε κάθε κλάση. Η μέθοδος επιτυγχάνει βελτιωμένη δικαιοσύνη και ακρίβεια κατά μήκος συνόλων δεδομένων (data sets) όπως τα MNIST, CIFAR-10, και CIFAR-100 σε σχέση με σύγχρονους (state-of-the-art) τεχνικές Δειγματοληψίας Ταμιευτήρα (Reservoir Sampling).Στο Κεφάλαιο 4 η διατριβή εξετάζει την Μάθηση Πολλαπλών Εργασιών στο πλαίσιο της Μέτα-Μάθησης (Meta-Learning). Η Μάθηση Πολλαπλών Εργασιών επιτρέπει τον ταυτόχρονο χειρισμό πολλαπλών εργασιών και προωθεί τον διαμοιρασμό πληροφορίας μεταξύ των εργασιών προκειμένου να ενισχυθεί η γενίκευση και η επίδοση, αλλά η Μέτα-Μάθηση ενισχύει περαιτέρω αυτές τις δυνατότητες εστιάζοντας στην προσαρμογή σε άπειρο πλήθος εργασιών, οι οποίες αντλώνται από κάποια κατανομή. Σε αυτό το πλαίσιο, η κωδικοποίηση εργασιών (task encoding) έχει ένα κεντρικό ρόλο: αναγνωρίζοντας την κωδικοποίηση της εργασίας, το μοντέλο αντιλαμβάνεται την εργασία που καλείται να αντιμετωπίζει και προσαρμόζεται ανάλογα. Μια κρίσιμη ιδέα σε αυτό το πλαίσιο είναι η μοντελοποίηση και η μάθηση μιας στοχαστικής απεικόνισης ή κωδικοποίησης και της κατανομής της έτσι ώστε να εμπεριέχει όσο περισσότερη πληροφορία γίνεται για κάθε εργασία και να οδηγεί σε ακριβής προβλέψεις. Αυτό είναι ιδιαίτερα σημαντικό σε περιβάλλοντα με μετακίνηση κατανομής (distribution drift), όπως αυτά της Συνεχούς Μάθησης, όπου οι εργασίες αλλάζουν με το χρόνο. Η διατριβή παρουσιάζει ένα πλαίσιο Μέτα-Μάθησης χρησιμοποιώντας τις έννοιες της Αμοιβαίας Πληροφορίας (Mutual Information) και της Αρχής Στενωπού της Πληροφορίας (Information Bottleneck Principle). Το πλαίσιο επιτρέπει στοχαστικές γενικεύσεις του γνωστού Αλγορίθμου Model-Agnostic Meta Learning (MAML) καθώς και νέους αλγορίθμους που βασίζονται στη μνήμη (Memory-based) οι οποίες εφαρμόζουν τεχνικές πυρήνων γκαουσιανών διαδικασιών, που επιτυγχάνουν ανταγωνιστική επίδοση σε σχέση με το state-of-the-art σε προβλήματα παρεμβολής και κατηγοριοποίησης σύντομης εκπαίδευσης (few-shot).Τεκμήριο Fighting an unfair battle: unconventional defenses against Advanced Persistent ThreatsVirvilis-Kollitiris, Nikolaos; Athens University of Economics and Business, Department of Informatics; Gritzalis, DimitrisThe number and complexity of cyber-attacks has been increasing steadily in recent years. The major players in today’s cyber conflicts are well organized and heavily funded teams with specific goals and objectives, working for or supported by a nation-state. A commonly used term to describe such teams/groups is Advanced Persistent Threat (APT). APT target the communication and information systems of government, military and industrial organizations and are willing to use vast amounts of money, time and expertise to reach their goals. A clear indication of the level of sophistication of APT is their impressive arsenal. The complexity and capabilities of recently discovered malware used to facilitate such attacks are remarkable: Stuxnet, Duqu, Flame, Red October, MiniDuke and more recently Regin are examples of highly sophisticated malware, the development of which required skillful individuals – in some cases (e.g. Stuxnet) with expertise in multiple technology fields – as well as substantial financial resources. In addition, serious insider attacks have occurred that resulted in the publication of several thousand classified documents, highlighting the fact that even in sensitive institutions, the effectiveness of the existing security safeguards is insufficient. Advances in attacker sophistication have not been matched by similar defensive advances. The concept of keeping the internal, trusted network separated from the external, untrusted one (i.e. boundary protection) has become obsolete. The use of blacklists or signatures for attack detection is practically useless against sophisticated attackers. The security industry, having spent decades developing security products such as anti-malware solutions and intrusion-detection/prevention systems, refuses to admit the shortcomings of these products. It is not uncommon for security companies to advertise that their products can detect and stop APT, even though the same products have been unable to detect such attacks for several years. Furthermore, C-level executives fail to understand the need for more robust security mechanisms, as they believe that by following vendor recommendations and making significant investments in traditional security solutions, they will keep their organization secure. However reality has proven them wrong, over and over again. In order to defend against such sophisticated adversaries, it is necessary to redesign our defenses and develop technologies focused more on detection than prevention. The purpose of this thesis is to offer a comprehensive view of the APT problem by analyzing the most common techniques, tools and attack paths that attackers are using, and highlighting the shortcomings of current security solutions. The use of deception techniques for attack detection is one of the integral focal points of this thesis. Based on this concept, a novel APT detection model is proposed, implemented and evaluated. The evaluation results highlight the significant efficacy of the model in detecting sophisticated attacks, with a very low false positive rate.Τεκμήριο Increasing security and resilience in cyber-physical processes of critical infrastructures(26-09-2023) Δεδούσης, Παναγιώτης; Dedousis, Panagiotis; Athens University of Economics and Business, Department of Informatics; Apostolopoulos, Theodoros; Stamatiou, Υannis; Mavridis, Ioannis; Katos, Vasilios; Kotzanikolaou, Panagiotis; Stergiopoulos, George; Gritzalis, DimitriosΗ σύγχρονη κοινωνία και η εθνική ευημερία εξαρτώνται σε μεγάλο βαθμό από τις Κρίσιμες Υποδομές (ΚΥ). Ωστόσο, σημαντικά ζητήματα ασφάλειας και αξιοπιστίας αποτελούν τροχοπέδη στη λειτουργία τους. Αυτή η διατριβή επικεντρώνεται στη δημιουργία νέων και καινοτόμων εργαλείων και μεθοδολογιών για την ανάλυση της επικινδυνότητας σε δίκτυα γράφων με σκοπό τη βελτίωση της Ασφάλειας και της Ανθεκτικότητας των ΚΥ.Αρχικά, προτείναμε μια μέθοδο που αυτοματοποιεί την ανάλυση των αλληλεξαρτήσεων αγαθών που υποστηρίζουν επιχειρηματικές διαδικασίες σε δίκτυα ΚΥ με σκοπό τη μείωση της επικινδυνότητας απέναντι σε κυβερνοαπειλές. Επιπλέον, προτείναμε μια μέθοδο για την αυτόματη ανάλυση σύνθετων γράφων επιθέσεων σε υποδομές νέφους, που μπορεί να εντοπίζει κρίσιμες ευπάθειες για το υπό μελέτη σύστημα.Αναπτύξαμε μια αυτοματοποιημένη μέθοδο που αξιοποιεί την εξόρυξη δεδομένων από αρχεία καταγραφής συμβάντων ενός πληροφοριακού συστήματος για τη διεξαγωγή ταχύτερων και πιο εμπεριστατωμένων Αποτιμήσεων Επικινδυνότητας. Η αποτελεσματική παρακολούθηση των επιφανειών επίθεσης των ΚΥ είναι ζωτικής σημασίας. Στο πλαίσιο αυτό, η προσέγγισή μας για τη διαδικασία εξόρυξης δεδομένων αναγνωρίζει απροσδόκητους διανυσματικούς επιθετικούς τομείς εντός των επιχειρηματικών διαδικασιών. Επιπλέον, αναπτύξαμε μια τεχνική δικτυακής επίθεσης (bit-masking) που επιτρέπει την κακόβουλη αμφίδρομη επικοινωνία, η οποία αποκαλύπτει ελαττώματα ασφαλείας στις λύσεις ασφαλείας και υπογραμμίζει την ανάγκη για προηγμένη πρόληψη των απειλών.Τα δίκτυα αεροπορικών μεταφορών παίζουν κρίσιμο ρόλο στις λειτουργίες των ΚΥ. Προς αυτήν την κατεύθυνση, παρουσιάσαμε μια μέθοδο βασισμένη στην επικινδυνότητα για την ανάλυση των αλληλεξαρτήσεων και των συμφορήσεων στο αεροπορικό δίκτυο, με σκοπό την διασφάλιση της ανθεκτικότητας των δικτύων αεροπορίας.Για να ενισχύσουμε την ανθεκτικότητα των Κυβερνο-Φυσικών Συστημάτων, προτείναμε ένα πλαίσιο που ενσωματώνει αρχές ασφάλειας και ανθεκτικότητας στον σχεδιασμό βιομηχανικών συστημάτων. Τέλος, παρουσιάσαμε μια μέθοδο που συνδυάζει και εφαρμόζει τις έννοιες των Digital twins και του Chaos Engineering σε βιομηχανικές εγκαταστάσεις και τις υποκείμενες διαδικασίες τους για τη δημιουργία μοντέλων με στόχο τη μελέτη και τη βελτίωση της ανθεκτικότητάς τους.Τεκμήριο Linear and non linear dimensionality reduction for distributed knowledge discoveryMagdalinos, Panagis; Athens University of Economics and Business, Department of Informatics; Vazirgiannis, MichalisAn increasing number of contemporary applications produce massive volumes of very high dimensional data. In scientific databases, for example, it is common to encounter large sets of observations, represented by hundreds or even thousands of coordinates. Unfortunately the rate of data generation and accumulation significantly outperforms our ability to explore and analyze it. Nevertheless, in order to extract knowledge from these datasets, we need to access the underlying, hidden information. However, the size and complexity of these collections makes their processing and analysis impractical or even ineffective [13, 47]. Therefore, scaling up knowledge discovery algorithms for data of both high dimensionality and cardinality has been recently recognized as one of the top-10 problems in data mining research [95].In parallel, the evolution of the internet as well as the emergence of novel applications, such as peer-to-peer systems, has led to an unprecedented distribution of available information. Data is dispersed among network nodes, making the cost of centralizing and subsequent processing prohibitive. Consequently, distributed data mining and distributed knowledge discovery have also emerged as highly challenging tasks[95]. Nevertheless, the vast amount of generated data dictates methods that are fast, exhibit low requirements in terms of computational resources and can be applied to various network setups. Motivated by the previous analysis, this thesis attempts to provide a solution through the definition of efficient and effective dimensionality reduction algorithms. The proposed methods exhibit minor requirements in terms of computational resources without compromising the quality of the produced results; therefore can be exploited in the context of centralized and distributed preprocessing for knowledge discovery. Towards this end,• We introduce FEDRA (Chapter 3, [62, 63]), a dimensionality reduction algorithm which poses minimum time and space requirements and is ideal for large FEDRA is an acronym for A Fast and Efficient Dimensionality Reduction Algorithm datasets of particularly high cardinality and dimensionality.• Inspired by the nature of landmark based dimensionality reduction algorithms(Chapter 2) we introduce the distributed adaptation of FEDRA ([62, 61]) and extend its underlying methodology in order to derive a framework for the decentralization of any landmark based dimensionality reduction algorithm (Chapter 3, Section 3.4)• We propose a distributed non linear dimensionality reduction algorithm, the Distributed Isomap (Chapter 4, [66, 65]) which to the best of our knowledge comprises the first of this kind. Additionally, motivated by recent research results on text-mining ([41, 17, 101, 78, 71]) we propose its application for hard dimensionality reduction problems related with text-mining.• Finally, we introduce X-SDR (Chapter 5, [64]), a prototype that enables the integration and evaluation of any dimensionality reduction algorithm. X-SDRis an open source tool that supports the evaluation of methods through experimentation on artificial and real world datasets thus promoting itself as an ideal candidate platform for research and teaching in academia.Τεκμήριο Marketing communication strategies and consumer response to corporate social irresponsibility: the role of tight and loose cultures(20-10-2023) Παπαδημητρίου, Ευαγγελία; Papadimitriou, Evangelia; Athens University of Economics and Business, Department of Informatics; Leonidou, Constantinos; Argouslidis, Paraskevas; Kokkinaki, Flora; Siomkos, Georgios; Tsogas, Markos; Baltas, George; Skarmeas, DionysiosΑυτή η διατριβή υποστηρίζει ότι οι καταναλωτές διαφέρουν ως προς την απόκριση τους σε περιστατικά Εταιρικής Κοινωνικής Ανευθυνότητας (ΕΚΑ) συναρτήσει της αυστηρότητας – χαλαρότητας της κουλτούρας. Η αυστηρότητα – χαλαρότητα της κουλτούρας αναφέρεται στη διακύμανση της ισχύος του κανόνα και της ανοχής για απόκλιση από τον κανόνα μεταξύ των κουλτούρων. Συγκεκριμένα, σε περίπτωση περιστατικού ΕΚΑ, οι καταναλωτές με πιο αυστηρή κουλτούρα αντιλαμβάνονται την εταιρεία ως λιγότερο ηθική, προκαλώντας αυξημένα αρνητικά ηθικά συναισθήματα. Κατά συνέπεια, αυτό οδηγεί σε λιγότερο ευνοϊκές αντιδράσεις των καταναλωτών προς την εταιρεία. Μέσω μίας σειράς από πέντε μελέτες, αυτή η διατριβή παρέχει συγκλίνουσες αποδείξεις για το φαινόμενο της αυστηρότητας – χαλαρότητας και τον υποκείμενο μηχανισμό μέσω της αντιληπτής ηθικής. Επιπροσθέτως, προσδιορίζει τις στρατηγικές επικοινωνίας μάρκετινγκ ως θεωρητικά και εμπειρικά σημαντικούς ρυθμιστές. Ειδικότερα, το φαινόμενο της αυστηρότητας – χαλαρότητας εμφανίζεται όταν η εταιρεία είναι είτε μη ανθρωπομορφοποιημένη είτε στερείται ζεστασιάς, προκαλώντας λιγότερη αντιληπτή ηθική στους καταναλωτές με πιο αυστηρές κουλτούρες. Ωστόσο, μετριάζεται όταν η εταιρεία είναι είτε ανθρωπομορφοποιημένη είτε υψηλή σε ζεστασιά. Επιπλέον, η επιρροή της αυστηρότητας – χαλαρότητας στην απόκριση των καταναλωτών προς την εταιρεία εξασθενεί όταν η εταιρεία χρησιμοποιεί την συγγνώμη και την αποζημίωση ως στρατηγική αντιμετώπισης κρίσεων. Τα ευρήματα των προαναφερθέντων μελετών υπογραμμίζουν ότι οι αντιδράσεις των καταναλωτών σε περιστατικά ΕΚΑ εξαρτώνται από την κουλτούρα καθώς και τις στρατηγικές επικοινωνίας μάρκετινγκ. Ως εκ τούτου, αυτή η έρευνα παρέχει πρακτικές πληροφορίες για την διαχείριση κρίσεων.Τεκμήριο Mechanisms and policies for resource allocation in clouds and next generation networks(02/24/2022) Darzanos, George; Νταρζάνος, Γεώργιος; Athens University of Economics and Business, Department of Informatics; Koutsopoulos, Iordanis; Courcoubetis, Costas A.; Polyzos, George C.; Siris, Vasilios A.; Dimakis, Antonios; Denazis, Spyros; Stamoulis, George D.Το οικοσύστημα του Διαδικτύου σήμερα αποτελείται από ένα μεγάλο εύρος φορέων όπως οι Πάροχοι Δικτύων Κινητής Τηλεφωνίας (Mobile Network Operators - MNOs), οι Πάροχοι Υπηρεσιών Υπολογιστικής Νέφους (Cloud Service Providers - CSPs), τα Δίκτυα Διανομής Περιεχομένου (Content Delivery Networks - CDNs), οι Πάροχοι Περιεχομένου (Content Providers - CPs), οι Πάροχοι Υπηρεσιών Over-the-Top (OTT) κ.λ.π.. Οι φορείς αυτοί προσφέρουν αμοιβαίως συμπληρωματικές λειτουργίες και αλληλεπιδρούν μεταξύ τους για την από κοινού παροχή υπηρεσιών προστιθέμενης αξίας σε επιχειρήσεις και τελικούς χρήστες. Ωστόσο, αυτοί οι φορείς είναι ορθολογικές (rational) οντότητες που αναζητούν κέρδη. Συνεπώς, αν και συνεργάζονται λειτουργικά, ανταγωνίζονται μεταξύ τους για τη διανομή των εσοδών από τους πελάτες των προσφερόμενων υπηρεσιών. Στην παρούσα διδακτορική διατριβή, προτείνουμε και αξιολογούμε υποδείγματα και μηχανισμούς που διευκολύνουν την αλληλεπίδραση αυτών των φορέων, καθώς και μηχανισμούς και πολιτικές που επιτυγχάνουν αποτελεσματική κατανομή πόρων και εσόδων σε συνεργατικά και σε ανταγωνιστικά οικοσυστήματα, καθώς και σε ενδιάμεσες περιπτώσεις συνεργασίας. Ειδικότερα, μελετάμε: (i) τη δημιουργία ομοσπονδιών από πολλαπλούς CSPs, εστιάζοντας στην παροχή υπηρεσιών υπολογιστικής νέφους που αξιοποιούν τους συλλογικά διαθέσιμους υπολογιστικούς πόρους, (ii) τις αλληλεπιδράσεις πολλαπλών MNOs κατά τη δημιουργία τεμαχίων δικτύου (network slices) τα οποία πρέπει να διασχίζουν πολλούς MNOs προκειμένου να καταστεί δυνατή η εξυπηρέτηση των αντίστοιχων από-άκρο-σε-άκρο εφαρμογών, και (iii) την από κοινού εξέταση της ανάθεσης χρηστών και της τοποθέτησης περιεχομένου σε Σταθμούς Βάσης Μικρού Κελιού (Small-cell Base Station - SBS), όπου ένας MNO και ένας CP συνεργάζονται για την αποτελεσματική διαχείριση των διαθέσιμων πόρων του ασύρματου δικτύου και της χωρητικότητας αποθήκευσης περιεχομένου (caches).
- «
- 1 (current)
- 2
- 3
- »