Dipartimento di Matematica e Informatica - Tesi di Dottorato

Permanent URI for this collectionhttps://lisa.unical.it/handle/10955/103

Questa collezione raccoglie le Tesi di Dottorato afferenti al Dipartimento di Matematica e Informatica dell'Università della Calabria.

Browse

Search Results

Now showing 1 - 10 of 13

Balancing the average weighted completion times of two classes of jobs: a new scheduling problem
(Università della Calabria, 2023-11-29) Avolio, Matteo; Terracina, Giorgio; Fuduli, Antonio
Exploring a new area of the scheduling theory and inspired by a real application in an academic context, in this thesis we introduce a new single-machine two-agent scheduling problem, aimed at balancing the average weighted completion times of two different classes of jobs, one per agent. Differently from the common multiagent cases, which are generally of the competing type, this problem could be interpreted as a cooperative type problem. In fact, even if the two agents share the same machine, they cooperate to optimize the unique global objective function, in order to balance their average weighted completion times. While for the case with identical jobs and unitary weight we present an exact algorithm providing an optimal solution in linear time, for the general case we prove the NP-hardness of the problem and we propose a mathematical formulation as a variant of the well known quadratic assignment problem. By applying the Glover linearization, we obtain a mixed integer linear program exploited to design a Lagrangian heuristics based on solving, at each iteration, a linear assignment problem. Since the proposed algorithm has revealed to be able to solve instances up to 500 jobs, in order to face larger scale instances (up to 2000 jobs) we also propose a genetic algorithm.
Expanding the Frontiers in GenAI and XAI: Innovative Architectures and Applications
(Università della Calabria, 2024-09-27) Adorneto, Carlo; Greco, Gianluigi; Terracina, Giorgio
Generative Artificial Intelligence (GenAI) and Explainable Artificial Intelligence (XAI) have attracted significant interest in recent years due to their potential and their capacity to drive and inspire further research. This thesis explores new frontiers in these fields by presenting a collection of innovative architectures and applications. In the realm of GenAI, this work introduces GIDnets, a generative neural network aimed at solving inverse design problems through latent space exploration, showcasing improvements over existing methods. Furthermore, the research explores the application of latent space conditioning and transformers for automatic medical report generation. The thesis also investigates the role of generative agents, based on Large Language Models (LLMs), in agent-based modeling, offering insights into their validation and the emerging challenges. One of the notable challenges addressed is the complexity of opinion diffusion in social environments, highlighting its potential as a promising application scenario for generative agents. In the domain of XAI, this thesis illustrates the impact of computational methods on data interpretation, particularly when data science and Deep Learning (DL) are employed to gain insights in the biomedical field. Despite advancements, explaining DL models remains a debated issue. SHAP (SHapley Additive exPlanations) is demonstrated as a powerful tool for extracting insights from these black-box models and its application in bankruptcy prediction and natural disaster event scenarios will be discussed. Additionally, a new deep learning algorithm based on XAI is proposed for feature selection in genomics. This algorithm utilizes a new SHAP-inspired metric to identify and quantify the impact of genes, significantly enhancing the prediction accuracy for chronic lymphocytic leukemia. The innovative approaches presented in this thesis advance the state-of-the-art in GenAI and XAI, showcasing the potential of these technologies to enable the design of practical solutions across various domains.
Optimizing Evaluation of Logic Programs: Extended Compilation & Enhanced Rewritings
(Università della Calabria, 2023-11-29) Mazzotta, Giuseppe; Terracina, Giorgio; Ricca, Francesco
SPARQL-QA: A system for Question Answering over RDF(S) Knowledge Bases
(Università della Calabria, 2023-04-29) Borroto Santana, Manuel Alejandro; Terracina, Giorgio; Ricca, Francesco
Enhancing DLV for reasoning over streams: the LDSR language and its expressiveness
(Università della Calabria, 2024-04-09) Morelli, Maria Concetta; Terracina, Giorgio; Manna, Marco; Perri, Simona
Crowdshipping in dynamic pickup and delivery problems
(Università della Calabria, 2023-11-23) Stoia, Sara; Terracina, Giorgio; Laganà, Demetrio; Vocaturo, Francesca
Deep Learning and Graph Theory for Brain Connectivity Analysis in Multiple Sclerosis
(Università della Calabria, 2020-01-16) Marzullo, Aldo; Leone, Nicola; Calimeri, Francesco; Terracina, Giorgio;
Multiple sclerosis (MS) is a chronic disease of the central nervous system, leading cause of nontraumatic disability in young adults. MS is characterized by inflammation, demyelination and neurodegenrative pathological processes which cause a wide range of symptoms, including cognitive deficits and irreversible disability. Concerning the diagnosis of the disease, the introduction of Magnetic Resonance Imaging (MRI) has constituted an important revolution in the last 30 years. Furthermore, advanced MRI techniques, such as brain volumetry, magnetization transfer imaging (MTI) and diffusion-tensor imaging (DTI) are nowadays the main tools for detecting alterations outside visible brain lesions and contributed to our understanding of the pathological mechanisms occurring in normal appearing white matter. In particular, new approaches based on the representation of MR images of the brain as graph have been used to study and quantify damages in the brain white matter network, achieving promising results. In the last decade, novel deep learning based approaches have been used for studying social networks, and recently opened new perspectives in neuroscience for the study of functional and structural brain connectivity. Due to their effectiveness in analyzing large amount of data, detecting latent patterns and establishing functional relationships between input and output, these artificial intelligence techniques have gained particular attention in the scientific community and is nowadays widely applied in many context, including computer vision, speech recognition, medical diagnosis, among others. In this work, deep learning methods were developed to support biomedical image analysis, in particular for the classification and the characterization of MS patients based on structural connectivity information. Graph theory, indeed, constitutes a sensitive tool to analyze the brain networks and can be combined with novel deep learning techniques to detect latent structural properties useful to investigate the progression of the disease. In the first part of this manuscript, an overview of the state of the art will be given. We will focus our analysis on studies showing the interest of DTI for WM characterization in MS. An overview of the main deep learning techniques will be also provided, along with examples of application in the biomedical domain. In a second part, two deep learning approaches will be proposed, for the generation of new, unseen, MRI slices of the human brain and for the automatic detection of the optic disc in retinal fundus images. In the third part, graph-based deep learning techniques will be applied to the study of brain structural connectivity of MS patients. Graph Neural Network methods to classify MS patients in their respective clinical profiles were proposed with particular attention to the model interpretation, the identification of potentially relevant brain substructures, and to the investigation of the importance of local graph-derived metrics for the classification task. Semisupervised and unsupervised approaches were also investigated with the aim of reducing the human intervention in the pipeline.
Generalizing identity-based string similarity metrics: theory and applications
(2018-01-19) Cauteruccio, Francesco; Leone, Nicola; Terracina, Giorgio
Le stringhe giocano un ruolo fondamentale in informatica: codificando i dati, la loro interpretazione permette di derivare informazione. Dato un insieme di stringhe, alcune interessanti domande emergono: “queste stringhe sono correlate?”, e se lo sono, “possiamo misura la loro correlazione?”. La definizione di un grado di similarit`a tra stringhe risulta essere fortemente importante. Varie definizioni di similarit`a tra stringhe sono state definite nella letteratura, derivanti dal concetto di metrica in matematica. Una delle pi`u famose metriche di similarit`a tra stringhe `e la edit distance, definita come il numero minimo di edit operation necessarie a trasformare una stringa in un’altra. Tuttavia, le varie definizioni presentano un’assunzione chiave: simboli uguali tra le stringhe rappresentano la stessa identica informazione, mentre simboli diversi introducono una qualche di↵erenza. Questa assunzione risulta essere estremamente riduttiva: esistono casi in cui l’identit`a tra simboli sembra non essere sufficiente a definire una similarit`a, e nel caso in cui non ci siano simboli in comune tra due stringhe, si pu`o verificare che simboli diversi rappresentino la stessa informazione. Inoltre, in alcuni casi una mappatura one-to-one tra i simboli risulta inefficace, quindi si necessita una mappatura many-to-many. La necessit`a di avere una metrica di similarit`a tra stringhe che sia in grado di catturare correlazioni nascoste tra le stringhe emerge, ove il concetto chiave `e rappresentato dal considerare che simboli di↵erenti possono esprimere concetti simili. Lo scopo di questa tesi `e di contribuire in questo scenario. In primis, un framework che generalizza la maggior parte delle metriche di similarit`a tra stringhe (basate sull’identit`a tra simboli) viene presentato, idoneo a scenari di applicazione in cui sono presenti stringhe definite su alfabeti eterogenei. La Multi-Parameterized Edit Distance (una generalizzazione della edit distance con il supporto del framework) viene definita formalmente e studiata dal punto di vista della complessit`a computazionale. In seguito, di↵erenti euristiche, definite, implementate e testate, vengono presentate, in modo da approcciarsi alle difficolt`a computazionali presenti. Varie euristiche sono presentate e tre di esse sono studiate, discusse e testate in dettaglio. Alcuni contesti di applicazione, studiati in questa tesi, sono quindi discussi, spaziando dal settore ingegneristico a quello informatico biomedico: anomaly detection nelle Wireless Sensors Area Network, analisi dei White Matter Fiber-Bundles e analisi degli Elettroencefalogrammi. Le conclusioni e una panoramica dei lavori futuri chiudono la tesi.
Answer set programming:development tols and applications to tourism
(2015-12-15) Nardi, Barbara; Leone, Nicola; Terracina, Giorgio
Answer Set Programming (ASP) is a declarative rule–based programming paradigm for knowledge representation and declarative problem–solving. The idea of ASP is to represent a given computational problem by using a logic program, i.e., a set of logic rule, such that its answer sets correspond to solutions, and then, use an answer set solver to find such solutions. Logic programming paradigms have received renewed interest in recent years, as demonstrated by emerging applications in many different areas of computer science, as well as industry. Due to this renewed interest an increased level of activity in the area has been registered which involved new partitioners both from academia and industry. The development of such applications has provided important information on the real potentials of this programming paradigm, especially concerning the capability of solving complex problems in practice; moreover, application developers highlighted some critical issues to be addressed to make ASP more effective and easy to use ASP in real-world. This thesis offers several contributions in this context can be summarized as follows: (i) The development of two applications of ASP in a specific industrial field; (ii) The design and implementation of new development tools for ASP. Concerning point (i), the thesis addresses two issues considered relevant in the tourism industry. The first is known in the literature as the problem of (semi- )automatic allotment of package tours; and the second is the intelligent management of personalized newsletters for customers of travel agency. The ASP-based solutions presented in the thesis confirm that ASP is an effective tool for solving complex real-world problems. Concerning point (ii), the thesis describes two new development tools that extend ASPIDE, a well-known integrated development environment for ASP. The first tool aims at making easier the writing of logic programs for novice programmers and is particularly suitable for those who prefer visual programming tools. In particular, the user can “ draw ” an ASP program composing graphically the logic rules. The second development tool described in the thesis answers a need arising in the scientific communities that study the usage of logic programming, and its extensions, for reasoning and querying ontologies. The goal is to integrate editing tools for ontologies with tools for the development/generation of logic programs. To this end, the thesis proposes a tool that connects two well-known development environments in the two fields,ASPIDE and Prot´eg´e, in an integrated environment. The main contributions presented in this thesis have been published in the following research papers: Barbara Nardi, Kristian Reale, Francesco Ricca, Giorgio Terracina: An Integrated Environment for Reasoning over Ontologies via Logic Programming. Web Reasoning and Rule Systems - 7th International Conference, RR 2013, Mannheim, Germany, July 27-29, 2013. (LNCS – Vol. 7994 – Springer – Pg. 253-258). Barbara Nardi: A Visual Syntax for Answer Set Programming. Web Reasoning and Rule Systems - 8th International Conference, RR 2014, Athens, Greece, September 15-17, 2014. (LNCS – Vol. 8741 – Springer – Pg.249- 250). Carmine Dodaro, Nicola Leone, Barbara Nardi, Francesco Ricca: Allotment Problem in Travel Industry: A Solution Based on ASP. Web Reasoning and Rule Systems - 9th International Conference, RR 2015, Berlin, Germany, August 4-5, 2015. (LNCS – Vol. 9209 – Springer – Pg. 77-92).
Tecniche per la valutazione distribuita di programmi logici
(2014-11-30) Barilaro, Rosamaria; Leone, Nicola; Terracina, Giorgio
Recent developments in IT, and in particular the expansion of networking technologies, have made quite common the availability of software architectures where data sources are distributed across multiple (physically-di erent) sites. As a consequence, the number of applications requiring to e ciently query and reason on natively distributed data is constantly growing. In this thesis we focus on the context in which it is necessary to combine data natively resides on di erent, autonomous and distributed sources and it is appropriate to deal with reasoning task to extract knowledge from the data, via deductive database techniques [1]. The aim is distributed evaluation of logic programs through an optimization strategy that minimizes the cost of the local process and data transmission. We considered that a single logic rule can be seen as a conjunctive query (possibly with negation), whose result must be stored in the head predicate. Then, starting from the conjunctive query optimization techniques, the idea is to extend the best results of these to evaluation of logic programs. In this context the methods based on structural analysis of the queries seem particularly promising. Indeed, logical rules often contain multiple interactions interactions among join variables [2]. In the case of simple queries (acyclic) there are several algorithms that ensure execution time with a polynomial upper bound. Structural methods [3, 4, 5, 6] attempt to propagate the good results of acyclic queries to larger classes of these, whose structure is cyclic, but with a low \degree of cyclicity". The Hypertree Decomposition technique [6] appears to be the most powerful since generalizes strongly all other structural methods and guarantees improved response times for each class of queries. Decomposition can be interpreted as an execution plan for the query, which rst requires the evaluation of the join associated with each cluster, and then requires the processing of the resulting join tree using a bottom-up approach. We used a weighted extension of Hypertree Decomposition [7] that combine structural analysis with evaluation of relevant quantitative information about the data, such as the number of tuples in relations, the selectivity of attributes and so on, and calculates minimum decompositions w.r.t. a cost function. We suitably modi ed this method in order to estimate the cost of data transmission between di erent sites resulting from the distribution of the sources and the correct evaluation of negation in rule bodies. According decomposition the query is transformed into a (tree-like) set of sub-queries which also allows the parallel evaluation of independent sub-query. We used parallel techniques combined with techniques for query optimization. We have adopted DLVDB [8, 9, 10, 11] as core reasoning engine, which allows to evaluate logic programs directly on database and combines appropriately expressive power of logic programming systems and the e cient data management of DBMSs. The interaction with databases is achieved by means ODBC connections, therefore, in case of distributed computing on network it allows to transparently access di erent sources and to express very simply distributed queries. We have implemented a prototype that we used to conduct the experiments. The preliminary results are very encouraging and showed the validity of the approach.

Dipartimento di Matematica e Informatica - Tesi di Dottorato

Browse

Filters

Settings

Sort By

Results per page

Search Results