Auflistung Computer Science nach Titel
-
Mladenović, Miljana (Beograd , 2016)[more][less]
Zusammenfassung: The beginning of the new millennium was marked by huge development of social networks, internet technologies in the cloud and applications of artificial intelligence tools on the web. Extremely rapid growth in the number of articles on the Internet (blogs, e-commerce websites, forums, discussion groups, and systems for transmission of short messages, social networks and portals for publishing news) has increased the need for developing methods of rapid, comprehensive and accurate analysis of the text. Therefore, remarkable development of language technologies has enabled their applying in processes of document classification, document clustering, information retrieval, word sense disambiguation, text extraction, machine translation, computer speech recognition, natural language generation, sentiment analysis, etc. In computational linguistics, several different names for the area concerning processing of emotions in text are in use: sentiment classification, opinion mining, sentiment analysis, sentiment extraction. According to the nature and the methods used, sentiment analysis in text belongs to the field of computational linguistics that deals with the classification of text. In the process of analysing of emotions we generally speak of three kinds of text classification: • identification of subjectivity (opinion classification or subjectivity identification) used to divide texts into those that carry emotional content and those that only have factual content • sentiment classification (polarity identification) of texts that carry emotional content into those with positive and those with negative emotional content • determining the strength or intensity of emotional polarity (strength of orientation). In terms of the level at which the analysis of feelings is carried out, there are three methodologies: an analysis at the document level, at the sentence level and at the level of attributes. Standardized methods of text classification usually use machine learning methods or rule-based techniques. Sentiment analysis, as a specific type of classification of documents, also uses these methods. This doctoral thesis, whose main task is the analysis of emotions in text, presents research related to the sentiment classification of texts in Serbian language, using a probabilistic method of machine learning of multinomial logistic regression i.e. maximum entropy method. The aim of this research is to create the first comprehensive, flexible, modular system for sentiment analysis of Serbian language texts, with the help of digital resources such as: semantic networks, specialized lexicons and domain ontologies. This research is divided into two phases. The first phase is related to the development of methods and tools for detecting sentiment polarity of literal meaning of the text. In this part of the work, a new method of reducing the feature vector space for sentiment classification is proposed, implemented and evaluated. The proposed method for reduction is applied in the classification model of maximum entropy, and relies on the use of lexical-semantic network WordNet and a specialized sentiment lexicon. The proposed method consists of two successive processes. The first process is related to the expansion of feature vector space by the inflectional forms of features. The study has shown that usage of stemming in sentiment analysis as a standard method of reducing feature vector space in text classification, can lead to incomplete or incorrect sentiment-polarity feature labelling, and with the introduction of inflectional feature forms, this problem can be avoided. The paper shows that a feature vector space, increased due to the introduction of inflectional forms, can be successfully reduced using the other proposed procedure – semantic mapping of all predictors with the same sentiment-polarity into a small number of semantic classes. In this way, the feature vector space is reduced compared to the initial one, and it also retains the semantic precision. The second phase of the dissertation describes the design and implementation of formal ontologies of Serbian language rhetorical figures – the domain ontology and the task ontology. Usage of the task ontology in generating features representing figurative speech is presented. The research aim of the second phase is to recognize figurative speech to be used in improving of the existing set of predictors generated in the first phase of the research. The research results in this phase show that some classes of figures of speech can be recognized automatically. In the course of working on this dissertation, a software tool SAFOS (Sentiment Analysis Framework for Serbian), as an integrated system for sentiment classification of text in Serbian language, has been developed, implemented and statistically evaluated. Results of the research within the scope of this thesis are shown in papers (Mladenović & Mitrović, 2013; Mladenović & Mitrović, 2014; Mladenović, Mitrović & Krstev, 2014; Mladenović, Mitrović, Krstev & Vitas, 2015; Mladenović, Mitrović & Krstev, 2016). The dissertation consists of seven chapters with the following structure. Chapter 1 introduces and defines methods, resources and concepts used in the first phase of research: text classification, sentiment classification, machine learning, supervised machine learning, probabilistic supervised machine learning, and language models. At the end of the introductory section, the tasks and objectives of the research have been defined. Chapter 2 presents a mathematical model of text classification methods and classification of sentiment methods. A mathematical model of a probabilistic classification and an application of the probabilistic classification in regression models are presented. At the end of the chapter it is shown that the method using the mathematical model of maximum entropy, as one of the regression models, has been successfully applied to natural language processing tasks. Chapter 3 presents the lexical resources of the Serbian language and the methods and tools of their processing. Chapter 4 deals with the comprehensive research on the currently available types and methods of sentiment classification. It shows the current work and research in sentiment classification of texts. It also presents a comparative overview of research in sentiment classification of texts using the method of maximum entropy. Chapter 5 discusses the contribution of this thesis to methods of feature space reduction for maximum entropy classification. First, a feature space reduction method is analysed. A new feature space reduction method which improves sentiment classification is proposed. А mathematical model containing proposed method is defined. Learning and testing sets and lexical-semantic resources that are used in the proposed method are introduced. Chapter 5 also describes building and evaluation of a system for sentiment classification – SAFOS, which applies and evaluates the proposed method of a feature vector space reduction. The parameters and the functions of SAFOS are defined. Also, measures for evaluation of the system were discussed – precision, recall, F1-measure and accuracy. A description of the method for assessing the statistical significance of a system is given. Also, implementation of the statistical test in the system SAFOS is discussed. The chapter provides an overview of the presented experiments, results and evaluation of the system. Chapter 6 deals with methods of recognizing figurative speech which can improve sentiment classification. The notion of domain ontology is introduced, the role of rhetorical figures and domain ontology of rhetorical figures. The importance of figurative speech in the sentiment classification has been explored. The description of the construction and structure of the first domain ontology of rhetorical figures in Serbian language, RetFig.owl, is given. Also, the description of the construction and structure of the corresponding task ontology that contains rules for identification of some classes of rhetorical figures is given. At the end of this chapter, an overview of the performed experiments, results and evaluation of the SAFOS system plugin that improved the recognition of figurative speech is given. The final chapter of this study deals with the achievemnts, problems and disadvantages of the SAFOS system. The conclusion of this thesis points to the great technological, social, educational and scientific importance of the sentiment analysis and recognition of the figurative speech and gives some routes in further development of the SAFOS system. URI: http://hdl.handle.net/123456789/4422 Dateien zu dieser Ressource: 1
Mladenovic_Miljana.pdf ( 13.60Mb ) -
Marovac, Ulfeta (Beograd , 2015)[more][less]
Zusammenfassung: Proteins are signi cant biological macromolecules of polymeric nature (polypeptides), which contain amino acids and are basic structural units of each cell. Their contents include 20+3 amino acids and, as a consequence, they are presented in biological databases as sequences formed from 23 di erent characters. Proteins can be classi ed based on their primary structure, secondary structure, function etc. One of possible classi cations of proteins by their function is related to their contents in a certain cluster of ortholologous groups (COGs). This classi cation is based on the previous comparison of proteins by their similarities in their primary structures, which is most often a result of homology, i.e. their mutual (evolutionary) origin. COG database is obtained by comparison of the known and predicted proteins encoded in the completely sequenced prokaryotic (archaea and bacteria) genomes and their classi cation by orthology. The proteins are classi ed in 25 categories which can be ordered in three basic functional groups (the proteins responsible for: (1) information storage and processing; (2) cellular processes and signaling; and (3) metabolism), or in a group of poorly characterized proteins. Classi cation of proteins by their contents in certain COG category (euKaryote Orthologous Groups- KOG for eukaryotic organisms) is signi cant for better understanding of biological processes and various pathological conditions in people and other organisms. The dissertation proposed the model for classi cation of proteins in COG categories based on amino acid n-grams (sequences of n- length). The set of data contains protein sequences of genomes from 8 di erent taxonomic classes [TKL97] of bacteria (Aqui cales, Bacteroidia, Chlamydiales, Chlorobia, Chloro exia, Cytophagia, Deinococci, Prochlorales), which are known to have been classi ed by COG categories. The new method is presented, based on the generalized systems of Boolean equations, used for separation of n-grams characteristic for proteins of corresponding COG categories. The presented method signi cantly reduces the number of processed n-grams in comparison to previously used methods of n-gram analysis, iv thus more memory space is provided and less time for protein procession is necessary. The previously known methods for classi cation of proteins by functional categories compared each new protein (whose function had to be determined) to the set of all proteins which had already been classi ed by functions in order to determine the group which contained most similar proteins to the one which was to be classi ed. In relation to the previous, the advantage of the new method is in its avoidance of sequence-sequence comparison and in search for those patterns (n-grams, up to 10 long) in a protein which are characteristic of the corresponding COG category. The selected patterns are added to a corresponding COG category and describe sequences of certain length, which have previously appeared in that COG category only, not in the proteins of other COG categories. On the basis of the proposed method, the predictor for determination of the corresponding COG category for a new protein is implemented. Minimal precision of the prediction is one of the predictors arguments. During the test phase the constructed predictor shown excellent results, with maximal precision of 99% reached for some proteins. According to its properties and relatively simple construction, the proposed method can be applied in similar domains where the solution of problem is based on n-gram sequence analysis. URI: http://hdl.handle.net/123456789/4308 Dateien zu dieser Ressource: 1
phdUlfetaMarovac.pdf ( 7.954Mb ) -
Šandrih, Branislava (Beograd , 2020)[more][less]
Zusammenfassung: The main goal of this dissertation is to put different text classification tasks inthe same frame, by mapping the input data into the common vector space of linguisticattributes. Subsequently, several classification problems of great importance for naturallanguage processing are solved by applying the appropriate classification algorithms.The dissertation deals with the problem of validation of bilingual translation pairs, sothat the final goal is to construct a classifier which provides a substitute for human evalu-ation and which decides whether the pair is a proper translation between the appropriatelanguages by means of applying a variety of linguistic information and methods.In dictionaries it is useful to have a sentence that demonstrates use for a particular dictio-nary entry. This task is called the classification of good dictionary examples. In this thesis,a method is developed which automatically estimates whether an example is good or badfor a specific dictionary entry.Two cases of short message classification are also discussed in this dissertation. In thefirst case, classes are the authors of the messages, and the task is to assign each messageto its author from that fixed set. This task is called authorship identification. The otherobserved classification of short messages is called opinion mining, or sentiment analysis.Starting from the assumption that a short message carries a positive or negative attitudeabout a thing, or is purely informative, classes can be: positive, negative and neutral.These tasks are of great importance in the field of natural language processing and theproposed solutions are language-independent, based on machine learning methods: sup-port vector machines, decision trees and gradient boosting. For all of these tasks, ademonstration of the effectiveness of the proposed methods is shown on for the Serbianlanguage. URI: http://hdl.handle.net/123456789/5090 Dateien zu dieser Ressource: 1
BranislavaSandrihPhd_konacno.pdf ( 9.053Mb ) -
Zeković, Ana (Beograd , 2015)[more][less]
Zusammenfassung: A main focus of the paper is construction of new methods for defining diverse knot distance types - the distance of knots made by crossing changes (Gordian distance) and the distance among knots made by crossing smoothing (smoothing distance). Different ways of knots presentation are introduced, with objective to a mirror curve model. It is presented a purpose of the model, coding of knots, by using the model preferences, as well as introduction of a method to determinate a knots presented by the model and derived all the knots that could be placed to a nets dimensions p×q (p ≤ 4, q ≤ 4). Diverse knot notations are described into details, with a focus to Conway’s notation and its topological characteristics. As it is known, a present algorithms are based on an algebra of chain fractions, that are in close relation with a presentation of rational knots, which results in an absence of a huge number of non-rational knots, in an existing Gordian’s distance tables. The subject of the paper is an implementation of methods with bases on determination of new distances equal 1. The methods are based on a non-minimal presentation of rational and non-rational knots, generation of algorithms established on geometrical characteristics of Conway’s notation and a weighted graph search. The results are organized into Gordian’s distance knots tables up to 9 crossings, and have been enclosed with the paper. In order to append the table with knots having a bigger number of crossings, it has been suggested a method for extension of results for knot families. Using facts of relation among Gordian’s numbers and smoothing numbers, a new method for smoothing number determination is presented, and results in a form of lists for knots not having more then 11 crossings. In conjunction with Conway’s notation concept and the method, algorithms for a smoothing distance are generated. New results are organized in knot tables, up to 9 crossings, combined with previous results, and enclosed with the paper. A changes and smoothing to a knot crossing could be applied for modeling topoisomerase and recombinase actions of DNA chains. It is presented the method for studying changes introduced by the enzymes. A main contribution to the paper is the concept of Conways notation, used for all relevant results and methods, which led to introduction of a method for derivation a new knots in Conways notation by extending C-links. In a lack of an adequat pattern for an existing knot tables in DT-notation, there is usage of a structure based on topological knot concepts. It is proposed a method for knot classification based on Conways notation, tables of all knots with 13 crossings and alternated knots with 14 crossings has been generated and enclosed. The subject of the paper takes into consideration Bernhard-Jablan’s hypothesis for a determination of unknotting number using minimal knot diagrams. The determination is crucial in computation of diverse knot distances. The paper covers one of main problems in knot theory and contains a new method of knot minimization. The method is based on relevance of local and global minimization. 5 There are defined new terms such as a maximum and a mixed unknotting number. The knots that do not change a minimum crossing number, after only one crossing change are taken into consideration for the analyzes. Three classes of the knots are recognized, and called by authors . Kauffman’s knots, Zekovic knots and Taniyama’s knots. The most interesting conclusion correlated with Zekovic knots is that all derived Perko’s knots (for n ≤ 13 crossings) are actually Zekovic knots. Defining this class of knots provides opportunity to emphasize new definitions of specifis featured for well-known Perko’s knots. URI: http://hdl.handle.net/123456789/4255 Dateien zu dieser Ressource: 1
phdZekovicAna.pdf ( 5.246Mb ) -
Radosavljević, Jovan (Beograd , 2023)[more][less]
Zusammenfassung: Graph G = (V,E) is an ordered pair of set of nodes V and branches E. Order graph G is the number of nodes |V |, and its size is the number of branches |E|. Knots u, v ∈ V are adjacent if there is a branch uv ∈ E between them. Distance dist(u, v) nodes u and v G is the length of the shortest path from u to v. The diameter of the graph G is the largest distance dist(u, v) let two nodes in, v. They are discussed in the dissertation graphs of diameter 2. Intuitively, the notion that graphs are dia- meters 2 simple structures; however, they are known to be asymptotically close all graphs of diameter 2. That is why a narrower class is interesting — class D2C of critical graphs of diameter 2, i.e. graphs where the removal of any branches leads to an increase in diameter. In addition, a narrower class of pri- mitive D2C (PD2C) graphs, i.e. D2C graphs that do not have two nodes with the same set of neighbors. In the introductory chapter 2, the basic concepts, algorithms and dings used in the dissertation. They are presented in the following chapters original results regarding diameter graphs 2. Chapter 3 describes the procedure for obtaining a list of D2C graphs of order up to 13. With built-in parallelization, the creation of a list of D2C graphs of order up to 13 it lasted a month. This was a step forward, because previously there was a spi- around all graphs of diameter 2 lines up to 10. The obtained results were used for testing several known hypotheses about graphs of diameter 2. In chapter 4 it is shown that for every m ⩾ 3 a D2C graph containing cli- a ku of size m must have at least 2m nodes. At the same time, with accuracy up to isomorphism, there is exactly one graph of size 2m that contains a clique of characters m. Chapter 5 discusses PD2C graphs with the smallest number of branches. From list of all PD2C graphs of order n ⩽ 13 are selected PD2C graphs of size at most 2n − 4. Only three of the isolated graphs are of size 2n − 5, which is in accordance with the statement of the Erdes-Renji theorem about the lower bound for the size graphs of diameter 2 that do not contain a node adjacent to all other nodes (that limit is 2n − 5). PD2C graphs of size 2n − 4 rows up to 13 sorted are in three groups: • The first group belongs to the Z family, defined in the dissertation, which for each n ⩾ 6 contains exactly one PD2C graph of order n of size 2n − 4. • The second group consists of seven Hamiltonian PD2C graphs of order at most 9 of size 2n−4. In the dissertation it was proved that there is no such Hamil- tone graph of order greater than 11, i.e. that the seven graphs found are the only ones Hamiltonian PD2C graphs of size 2n − 4. • The third group consists of a unique graph that does not belong to any of the first two groups. Based on these results, the hypothesis was formulated that all PD2C graphs re- that n ⩾ 10 and sizes 2n − 4 belong to the family Z. Keywords: graphs, critical graphs of diameter 2, primitive graph- You Scientific field: Computing and informatics Narrower scientific field: Graph theory UDC number: 004.415.5(519.1 URI: http://hdl.handle.net/123456789/5594 Dateien zu dieser Ressource: 1
disertacijaJovanRadosavljevic.pdf ( 746.0Kb ) -
Spasić, Mirko (Beograd , 2020)[more][less]
Zusammenfassung: The query containment problem is a very important computer science problem,originally defined for relational queries. With the growing popularity of theSPARQLquerylanguage, it became relevant and important in this new context, too. This thesis introducesa new approach for solving this problem, based on a reduction to satisfiability in first orderlogic. The approach covers containment underRDF SCHEMAentailment regime, and it candeal with the subsumption relation, as a weaker form of containment. The thesis provessoundness and completeness of the approach for a wide range of language constructs. It alsodescribes an implementation of the approach as an open source solverSPECS. The experi-mental evaluation on relevant benchmarks shows thatSPECSis efficient and comparing tostate-of-the-art solvers, it gives more precise results in a shorter amount of time, while suppor-ting a larger fragment ofSPARQLconstructs. An application of query language modeling canbe useful also along refactoring of database driven applications, where simultaneous changesthat include both a query and a host language code are very common. These changes canpreserve the overall equivalence, without preserving equivalence of these two parts consideredseparately. Because of the ability to guarantee the absence of differences in behavior betweentwo versions of the code, tools that automatically verify code equivalence have great benefitsfor reliable software development. With this motivation, a custom first-order logic modelingof SQL queries is developed and described in the thesis. It enables an automated approachfor reasoning about equivalence ofC/C++programs with embedded SQL. The approach is implemented within a publicly available and open source framework SQLAV. URI: http://hdl.handle.net/123456789/5095 Dateien zu dieser Ressource: 1
doktoratMirkoSpasic.pdf ( 1.258Mb ) -
Dražić, Zorica (Beograd , 2014)[more][less]
Zusammenfassung: The Variable neighborhood search method proved to be very successful for solving discrete and continuous optimization problems. The basic idea is a systematic change of neighborhood structures in search for the better solution. For optimiza- tion of multiple variable functions, methods for obtaining the local minimum starting from certain initial point are used. In case when the continuous function has many local minima, nding the global minimum is usually not an easy task since the obta- ined local minima in most cases are not optimal. In typical implementations with bounded neighborhoods of various diameters it is not possible, from arbitrary point, to reach all points in solution space. Consequently, the strategy of using the nite number of neighborhoods is suitable for problems with solutions belonging to some known bounded subset of IRn. In order to overcome the previously mentioned limitation the new variant of the method is proposed, Gaussian Variable neighborhood search method. Instead of de ning the sequence of di erent neighborhoods from which the random point will be chosen, all neighborhoods coincide with the whole solution space, but with di e- rent probability distributions of Gaussian type. With this approach, from arbitrary point another more distant point is theoretically reachable, although with smaller probability. In basic version of Variable neighborhood search method one must de ne in advance the neighborhood structure system, their number and size, as well as the type of random distribution to be used for obtaining the random point from it. Gaussian Variable neighborhood search method has less parameters since all the neighborhoods are theoretically the same (equal to the solution space), and uses only one distribution family - Gaussian multivariate distribution with variable dispersion. File transfer scheduling problem (FTSP) is an optimization problem widely appli- cable to many areas such as Wide Area computer Networks (WAN), Local Area Ne- v tworks (LAN), telecommunications, multiprocessor scheduling in a MIMD machines, task assignments in companies, etc. As it belongs to the NP-hard class of problems, heuristic methods are usually used for solving this kind of problems. The problem is to minimize the overall time needed to transfer all les to their destinations for a given collection of various sized les in a computer network, i.e. to nd the le transfer schedule with minimal length. In order to obtain the exact solution of the FTS problem, integer linear pro- gramming formulations are proposed and their correctness is proved. In this way optimal solutions can be found for small and medium size test instances. For large test instances, the Variable neighborhood search method is proposed using the "permutation" representation and typical neighborhood structures. More- over, the same method is used for obtaining the upper bounds of the solutions which are used in proposed integer linear programming formulations. For obtaining be- tter solutions in the small neighborhood of the current solution, three di erent local search procedures are implemented: 2-swap, 2-swap adjacent and variable neighbo- rhood descent. In order to apply the continuous optimization methods for solving FTSP, the weighted solution representation is developed. Such representation enables the co- ntinuous optimization methods to be used, which do not require the di erentiability of objective function. Since Gauss Variable neighborhood search method proved to be successful in continuous optimization problems, it was applied to FTSP. Pre- viously described local search procedures can also be used with weighted solution representation. Using the proposed methods optimal solutions for all small and medium size test instances are found. For large size instances, which are beyond the reach of exact methods, metaheuristic methods obtained good solutions in reasonable time. URI: http://hdl.handle.net/123456789/4246 Dateien zu dieser Ressource: 1
phdDrazic_Zorica.pdf ( 4.739Mb ) -
Tanasijević, Ivana (Beograd , 2020)[more][less]
Zusammenfassung: The motivation for writing this doctoral dissertation is a multimedia col-lection that is the result of many years of field research conducted by researchers from the Institute for Balkan studies of the Serbian Academy of Sciences and Arts. The collection consists of materials in the form of recorded interviews, various recorded customs, associated textual descriptions (protocols) and numerous other documents.The subject of research of this dissertation is the study of possibilities and the development of new methods that could be used as a starting point in solving the problem of managing the intangible cultural heritage of the Balkans. The subtasksthat emerge in this endeavor are the development of adequate design and implemen-tation of a multimedia database of intangible cultural heritage that would meet the needs of different types of users, automatic semantic annotation of protocols using natural language processing methods, as a basis for semi-automatic annotation of the multimedia collection, and successful search by metadata which comply with the CIDOC CRM standard, study of additional search possibilities of this collection in order to gain new knowledge, as well as development of selected methods.The main problem with the available methods is that there is still not enough developed infrastructure in the context of natural language processing, organization and management in the field of cultural heritage in the Balkans and especially for the Serbian language, which could be effectively used to solve the proposed problem.There is thus a strong need to develop methods to reach an appropriate solution.For the semi-automatic annotation of multimedia materials, automatic semantic annotation of the protocols associated with the materials was used. It was carriedout by methods of information extraction, recognition of named entities and topicextraction, using rule-based techniques with the help of additional resources suchas electronic dictionaries, thesauri and vocabularies from a specific domain.To classify textual protocols in relation to the topic, research was conducted onmethods that can be used to solve the problem of classifying texts in the Serbianlanguage, and a method was offered that is adapted to the specific domain beingprocessed (intangible cultural heritage), to the specific problems being solved (clas-sification of protocols in relation to the topic) and to the Serbian language, as one of the morphologically rich languages.To work with spatial data, a spatial model has been developed that is suitable for displaying results on a map, as well as for creating spatial queries through an interactive graphical display of a map of locations.The results of experiments conducted on the developed methods show that the use of a rule-based approach in combination with additional language resources an dwith putting in a reasonable amount of effort gives very good results for the task of information extraction. An F measure of 0.87 was reached for the extraction of named entities, while an F measure of 0.90 was reached for the extraction of topics,which is in the range of measures from published research from similar problem sand domains.The results of the text classification indicate that the selected statistical methods of machine learning in their basic form when applied to the protocols, although generally successful, give a bad F measure, 0.44, while significant improvement is achieved with the use of semantic techniques, in which case an F measure of 0.88 is reached.Some of the results presented in this dissertation are contained in the papers[266], [265], [94], [264], [267], which have been published or accepted for publication.The conclusion drawn from the research is that to solve the given problem it is necessary to engage experts from several fields, that the needs of different groups of users are complex, which complicates the task of organizing and managing them ultimedia collection, that the domain of cultural heritage is very rich in semantics,that context plays a major role in the tasks of information extraction and text classification, and finally that for these tasks the developed rule-based methods of natural language processing as well as statistical techniques of machine learning prove to be successful. URI: http://hdl.handle.net/123456789/5093 Dateien zu dieser Ressource: 1
IvanaTanasijevicdr.pdf ( 4.674Mb ) -
Džamić, Dušan (Beograd , 2021)[more][less]
Zusammenfassung: The theory of complex networks has proven to be very important inthe study of the characteristics and structure of various complex systems. In thelast two decades, a large number of researches have been directed towards thedevelopment of methods for clustering in complex networks. In 2012, Center forDiscrete Mathematics and Theoretical Computer Science (DIMACS), which is awell-known consortium of prestigious academic institutions (Rutgers University,Princeton, Colombia) and research laboratories (Microsoft, IBM, AT & T, NEC),included the problem of clustering in complex networks on the list of the mostimportant problems and challenges in computer science. Clustering in complexnetworks can be applied in a variety of contexts to achieve different goals, andtherefore, there is no generally accepted definition of a cluster. For this reason,different approaches are used in developing clustering methods. An approach thathas attracted the most attention of researchers involves two subproblems: defininga function to determine the quality of a partition and constructing methods to finda partition that has the maximum value of the defined quality function. In thisapproach, the problem of clustering is formulated as the problem of combinatorialoptimization and various methods of mathematical optimization can be used tosolve it. One of the most commonly used quality function is the modularity.Clustering by modularity maximization, i.e., finding a partition with the max-imum value of modularity, is NP-hard problem. Thus, only heuristic algorithmsare suitable of processing large datasets. In this dissertation, a novel method formodularity maximization based on the variable neighborhood search heuristic isproposed. For the purpose of efficient application in large-scale complex networks,a procedure for decomposition of the problem into smaller subproblems is devel-oped. In addition, a mechanism for overcoming local maxima of modularity isimproved using criteria for occasional acceptance of solution which is worse thanthe current one. DIMACS instances are used to test the proposed method, and theobtained results are compared with the best ones presented in the literature, ob-tained by two methods developed in DIMACS implementation challenge in 2012.In addition, the obtained results are compared with the results of six methodsdeveloped after 2012, from the literature. A comparative analysis of the resultsshows that the proposed method outperforms the existing methods for modularitymaximization and improves the best known solutions on 9 out of 13 hard instances. Clustering by modularity maximization is not suitable for detecting small clus-ters in large networks. For this reason, a new function for measuring the qualityof a partition has been proposed in the dissertation. Through three theorems, itis shown that the new measure, called E-function, has the potential to identifyclusters in the network and overcome limits of modularity. For testing the pro-posed E-function and comparison with the modularity function, a generic variableneighborhood method is developed to optimize the considered quality function.Computational experiments are performed on generated and real instances fromthe literature for which the correct division into clusters is known. The resultsshow that the expected clusters can be identified, both on artificial and real in-stances, by maximizing the E-function. URI: http://hdl.handle.net/123456789/5214 Dateien zu dieser Ressource: 1
teza_dzamic.pdf ( 3.074Mb ) -
Vučković, Bojan (Beograd , 2017)[more][less]
Zusammenfassung: We present original results from the following fields of discrete mathematics: chromatic graph theory, extremal set theory and Boolean matrix theory. From the chromatic graph theory we investigate edge and total colorings satisfying the condition that neighboring vertices of a graph possess different values of multi-set, set or sum, induced by the giving coloring. Multi-set neighbor-distinguishing edge coloring of a graph is an assignment of colors to edges such that, for every edge uv of a graph, multi-set of the edges incident with the vertex u differs from the multi-set of the edges incident with the vertex v. The previous best result concerning the minimum number of colors required for such a coloring of an arbitrary graph states that four colors are sufficient. The author’s contribution is a proof that such a coloring is always possible with only three colors, which is in general case the optimal number of colors. We construct a graph for which we subsequently prove that a different number of colors is required to obtain a multi-set neighbor-distinguishing coloring and neighbor-distinguishing coloring by sum. As far as we know, this is the first example of such a graph. A few results concerning the neighbor expended sum distinguishing coloring are given. The main contribution is a proof that for an arbitrary graph there exists a total coloring from the set f1; 2; 3g, such that every two adjacent vertices have different sums of its adjacent vertices and incident edges. Also, for certain classes of graphs is proved that there exists such a coloring using only the colors from the set f1; 2g. Neighbor-distinguishing edge coloring of a graph G requires that every two adjacent edges receive different colors, while the sets of the edges incident with the vertices u and v differ for every edge uv of G. The author presents a procedure of edge coloring for an arbitrary graph without isolated edges, where we a smaller number of colors is used compared to all known results. For the adjacent vertex distinguishing total coloring of a graph G the condition is that every two adjacent and incident elements of V (G) [ E(G) receive different colors, while for every edge uv of G the set composed from the colors assigned to the edges incident with u together with the color of u, differs from such a set for v. The author improves the upper bound of the minimum number of colors needed for such a coloring, relative to the maximal degree of a graph. Frankl’s conjecture from the extremal set theory states that for every family closed under union there exists an element contained in at least half of the sets of the family. We give a proof that Frankl’s conjecture holds for every family contained from 12 elements, while it is known that this is true for families contained from 11 or less elements. Our proof is based on the efficient algorithm that exhausts all the possibilities, while using the results for subfamilies that eventual counter-example cannot contain, which we obtained in a number of consecutive steps. Family of sets G is an FC-family if for every family F containing G there exists an element from S G that appears in at least half of the sets of F. NonFC-family is every family that is not FC. The author’s contribution is the complete classification of all families consisting of 6 or less elements into FC and NonFC-families. From the Boolean matrices theory we present our results concerning the row space cardinality. Boolean matrices are the matrices whose all components are from the set f0; 1g, while the row space of a Boolean matrix is the set of vectors that can be obtained by disjunction from the rows of a matrix. We present the set consisted of all values a from the interval [2n2 + 2n3; 2n2] such that there exists a matrix of dimension n n having the row space cardinality equal to a. For the least positive integer an for which there exists no matrix of dimension n n having the row space cardinality equal to an, the author gives a lower bound that is an improvement over the previously known results. All proofs for the main results in the dissertation are constructive. Proofs of some of them require the use of computers where there is a calculation of a great number of possibilities. For other proofs this was not necessity, though algorithms following the steps of the proofs can be implemented to obtain a graph coloring or a matrix with the desired properties. URI: http://hdl.handle.net/123456789/4661 Dateien zu dieser Ressource: 1
Disertacija_-_Bojan_Vuckovic.pdf ( 1.143Mb ) -
Filipović, Vladimir (Beograd , 2006)[more][less]
URI: http://hdl.handle.net/123456789/2478 Dateien zu dieser Ressource: 1
phdFilipovicVladimir.pdf ( 2.145Mb ) -
Pejović, Aleksandar (University of Novi Sad, Faculty of Technical Sciences , 2020)[more][less]
Zusammenfassung: This dissertation is about the development of a parallel software system for representing and solving problems of finite model theory and its application. The theoretical foundation of the system is presented, as well as an in-depth explanation of the implementation in Python. In particular, a parallel method for computing Boolean expressions based on the properties of finite free Boolean algebras is developed. It is also shown how various finite combinatorial objects can be coded in the formalism of Boolean algebras and counted by this procedure. Specifically, using a translation of first order predicate formulas to propositional formulas, we developed a technique for constructing and counting finite models of first order theories. Finally, we have developed some general techniques that enable more effective use of our system. We illustrate these techniques on two examples. The first one deals with partial orders, while the other one is about random graphs. URI: http://hdl.handle.net/123456789/4857 Dateien zu dieser Ressource: 3
APejovicDissertation.pdf ( 1.253Mb )APejovicPhDPresentation.pdf ( 1.103Mb )APejovicPhDSoftwareSources.zip ( 25.25Kb ) -
Carić, Marko (Beograd , 2023)[more][less]
Zusammenfassung: This dissertation discusses the problem of calculating the number of equivalent classes of a Boolean function. The difficulty of determining the number of equivalence classes increases sharply with the number of variables sn. The motivation for choosing this topic lies in the fact that concrete numbers have been known so far only for relatively small values of n, although the problem itself has long been theoretically solved. Let G be the permutation group of the set Bn={0,1}n. Effect of Gonscalar group, Bn→B1, that is, vector invertible Boolean functions, Bn→Bn. Two scalar Boolean functions f(k)ig(k), defined by Bn, are considered equivalent. ∈G forever ∈Bnf(k )=g(σ(k)) holds. Two vector invertible Boolean functions f(k) and g(k) are considered equivalent in relation to the group G, i.e. ))).The equivalence relation∼decomposes the sets of all Boolean functions into equivalence classes.The equivalence of Boolean functions confers significant applications in the logical synthesis of combinatorial circuits and cryptography, especially in connection with the design of S-boxes. Let Un(G) and Vn(G) denote the number of scalar equivalence classes, ie. vector invertible Boolean functions of n variables with respect to the group G. The numbers Un(G) and Vn(G) can be calculated relatively simply if the cycle index from the group G to the G group is known. Induced by the group Snpermutations of coordinate elements step = (k1,k2,...,kn)∈Bn, •group Gn, induced by permutations and additions of coordinates, •group GLnlinear invertible transformations of elements of vector space Bn, and •group AGLof invertible transformations of elements Bn. If the permutation σ∈Ghasikcycles of lengthk1, its cycle structure is(σ)=(i1,i2,...).The cyclic index of the groupGisgeneratrix ZG(f1,f2,...)=1 |G|σ∈Gk1 fik k structure cycles σene structures of cycles σfraconstructionsGsides ∗pressuresGpermutations of the red group are known, but the cycle indexes itself, i.e. numbers Un(G) and Vn(G), are practically calculated only for relatively small values, for example n10. The dissertation presents original results in the field of enumeration of equivalence classes of Boolean functions in relation to these four groups of soft transformations. A similar expression is derived for all four groups of soft transformations for the cycle index in the form of the sumover partition softhennumbern. Based on the precyclical index much more. An overview of known results for relatively small and new results in the thesis for larger ones are shown in the following table: Number\GSnGnGLnAGLn Un(G)11→3310→328→3110→31 Vn(G)6→307→276→266, participates in direct effect, by special effect 26 latingthenumberofequivalenceclassesthatdoesnotuseacycleindex is shown and described in the third paper from the introductory chapter. These second parts of the dissertation refer to monotone Boolean functions — scalar Boolean functions that satisfy the condition of monotonicity (from which follows f(k)f(i)). Letrn, i.e. monotoneBooleanfunctionsofnvariables. The difficulty of calculating the number rn increases rapidly with n, so that until recently the last term of the sequence to be calculated was r7. The procedure described in the dissertation is based on Frobenius' theorem, on the basis of which the number r8 was determined. The dissertation consists of the first introductory chapter and the following three chapters. In the second chapter, theoretical concepts related to the material from chapters 3 and 4 are introduced, and they relate to discrete mathematics, combinatorics and the index of the cycle considered by the soft cycle of the 3rd form for describing the 3rd cycle. indices for the four considered groups of permutations, as well as the numbers Un(G) and Vn(G) of the equivalent class of the Boolean function with respect to these groups. First, the common improvements for all four groups are discussed, and then the specific accelerations related to individual groups are published by group. ter. In Chapter 4, the problem of finding the number of equivalence classes of a monotone Boolean function is solved. First, a general expression for calculating a number is given based on the Frobenius theorem of the form of the sum (by the soft number partition of the number) of the number from the fixed part of the graph corresponding to the point corresponding to the point. In different partitions, different ways of calculating the number of non-fixed points for n8 are shown. The procedure based on which the number r8 was calculated, which also represents the original contribution of this dissertation, is shown in the first paper from the list from pp. 1 to 3. practically at the same time the obtained result described in the dissertation. URI: http://hdl.handle.net/123456789/5571 Dateien zu dieser Ressource: 1
MarkoCaricDisertacija.pdf ( 1.467Mb ) -
Maljković, Mirjana (Beograd , 2021)[more][less]
Zusammenfassung: Proteins are linear biological polymers composed of amino acids whose structure and function are determined by the number and order of amino acids. The structure of the protein has three levels: primary, secondary and ter- tiary (three-dimensional, 3D) structure. Since the experimental determination of protein 3D structure is expensive and time-consuming, it is important to develop predictors of protein 3D structure properties from the amino acid sequence (pri- mary structure), such as 3D structure of the protein backbone. The 3D structure of the backbone can be described using prototypes of local protein structure, i.e. prototypes of protein fragments with a length of few amino acids. A set of local structure prototypes determines the library of local protein structures, also called the structural alphabet. A structural alphabet is defined as a set of N proto- types of L amino acid length. The subject of this dissertation is the development of models for the prediction of structural alphabet prototypes for a given amino acid sequence using different data mining approaches. As one of the most known, structural alphabet Protein Blocks (PBs) was used in one part of the doctorial re- search. Structural alphabet PBs consists of 16 prototypes that are defined using fragments of 5 consecutive amino acids. The amino acid sequence is combined with the structural properties of a protein that can be determined based on amino acid sequence (occurrence of repeats in the amino acid sequence) and results of predictors of protein structural properties (backbone angles, secondary structures, occurrence of disordered regions, accessible surface area of amino acids) as an input to the prediction model of structural alphabet prototypes. Besides the de- velopment of models for prediction of prototypes of existing structural alphabet, the analysis of the capability of developing new structural alphabets is researched by applying the TwoStep clustering algorithm and construction of models for the prediction of prototypes of new structural alphabets. Several structural alpha- bets, which differ in the length of prototypes and the number of prototypes, have been constructed and analyzed. Fragments of the large number of proteins, whose structure is experimentally determined, were used to construct the new structural alphabets. URI: http://hdl.handle.net/123456789/5236 Dateien zu dieser Ressource: 1
dthesis.Matf.Mirjana.Maljkovic.pdf ( 47.43Mb ) -
Radojčić, Nina (Beograd , 2018)[more][less]
Zusammenfassung: In this dissertation, three NP-hard optimization problems are studied and va- rious computational intelligence methods are considered for solving them, with a special emphasis on the possibilities of applying fuzzy logic in order to improve the performances of proposed methods. In addition, it is shown how fuzzy logic can be incorporated into a model to make it more adequate to real world applications. The first problem considered is the Risk-Constrained Cash-in-Transit Vehicle Routing Problem (RCTVRP) that represents a special case of the vehicle routing problem (VRP). Similar to the classical VRP, the aim is to determine the collection routes from one depot to a number of customers in order to minimize the overall travel distance (or cost). Additionally, the safety aspect of the routed risk constraints are introduced in the case of RCTVRP. The RCTVRP concerns the issue of secu- rity during the transportation of cash or valuable goods (e.g. in the cash-in-transit industry). The other two problems studied in this dissertation belong to the class of loca- tion problems: the Load Balancing Problem (LOBA) and the Max-Min Diversity Problem (MMDP). The goal of the LOBA problem is to locate a fixed number of facilities, such that the difference between the maximum and minimum number of customers served by each facility is balanced. The LOBA model is useful in cases where customers naturally choose the closest facility. The MMDP consists of se- lecting a subset of a fixed number of elements from a given set in such a way that the diversity among the selected elements is maximized. This problem also arises in real world situations encompassing a variety of fields, particularly the social and biological sciences. In order to solve the RCTVRP, a fuzzy GRASP (Greedy Randomized Adaptive Search Procedure) is hybridized with Path Reliking (PR) methodology. Carefully adjusted fuzzy modification incorporated into the proposed GRASP for the RC- TVRP improved its performance. Moreover, in this dissertation a new PR structure is implemented and can be used for other vehicle routing problems. To improve the algorithm’s time complexity, a new data structure for the RCTVRP is incor- porated. The proposed fuzzy GRASP with PR hybrid shows better computational performance compared to its non-fuzzy version. Furthermore, computational results on publicly available data sets indicate that the proposed algorithm outperforms all existing methods from the literature for solving the RCTVRP. For solving the LOBA problem two efficient hybrid metaheuristic methods are proposed: a combination of reduced and standard variable neighborhood search met- hods (RVNS-VNS) and hybridization of evolutionary algorithm and VNS approach (EA-VNS). The proposed hybrid methods are first benchmarked and compared to all the other methods on existing test instances for the LOBA problem with up to 100 customers and potential suppliers. In order to test the effectiveness of the pro- posed methods, we modify several large-scale instances from the literature with up to 402 customers and potential suppliers. Exhaustive computational experiments show that the proposed hybrid methods quickly reach all known optimal solutions while providing solutions on large-scale problem instances in short CPU times. Re- garding solution quality and running times, we conclude that the proposed EA-VNS approach outperforms other considered methods for solving the LOBA problem. EA approach is also proposed for solving the MMDP. Computational experi- ments on a smaller benchmark data set showed that the classic EA quickly reached all optimal solutions obtained previously by an exact solver. However, some of the larger instances of MMDP were challenging for classic EA. Although researchers have established the most commonly used parameter setting for EA that has good performance for most of the problems, it is still challenging to choose the adequate values for the parameters of the algorithm. One approach to overcome this is chan- ging parameter values during the algorithm run. As part of this dissertation this problem was addressed by extending the evolutionary algorithm by adding a fuzzy rule formulated from EA experts’ knowledge and experience. The implemented fuzzy rule changes the mutation parameter during the algorithm run. The results on tested instances indicate that the proposed fuzzy approach is more suitable for solving the MMDP than classic EA. For all three problems addressed whereas the smaller instances that CPLEX was able to solve, obtained optimal solutions were used for comparison with proposed methods and all of the proposed methods obtained these optimal solutions. Moreover, in this dissertation it has been shown that fuzzy logic is a successful tool in modeling the RCTVRP. In this problem the risk constraints are set by using a risk threshold T on each route and thus, the routes with risk larger than T are forbidden. However, in this dissertation the aim is to take into account the probability of being robbed along each route instead of just allowing solutions with routes that satisfy the risk constraints. A new fuzzy model for the RCTVRP is developed which takes into account the value of the risk index of each route and the solutions with lower values of risk indexes on their routes are considered superior. In order to achieve that fuzzy numbers are used in the improved model. Moreover, two mixed integer program formulations of new fuzzy model are developed and presented in this dissertation. The introduced fuzzy model is compared with the model from the literature using an adequate example and the advantages of the newly proposed fuzzy RCTVRP is demonstrated. Computational experiments are performed and the comparison of the two models given in the paper show that the newly presented approach leads to safer routes. URI: http://hdl.handle.net/123456789/4737 Dateien zu dieser Ressource: 1
tezaNinaRadojicic.pdf ( 1.665Mb ) -
Kartelj, Aleksandar (Beograd , 2014)[more][less]
Zusammenfassung: This work investigates the potential of improving the classi cation process through solving three classi cation-related problems: feature selection, feature weighting and parameter selection. All three problems are challenging and currently in the focus of scienti c researches in the eld of machine learning. Each problem is solved by using population-based metaheruistic method called electromagnetismlike method. This method is used for combinatorial and global optimization. It is inspired by laws of attraction and repulsion among charged particles. Each particle is represented by a vector of real values. The solution of the problem of interest is then obtained by mapping these real-valued vectors to the feasible solution domain. Particles representing better solutions achieve higher level of charge, which consequently produces greater impact on other particles. The search process is performed by iterating the particle movement, induced by charges. Through implementing the methods, two key aspects are managed: 1) the classi cation quality obtained after applying the optimization method and 2) the e ciency of the proposed methods from the perspective of time and space resources. All methods are equiped with problem-speci c local search procedures which tend to increase the solution quality. The bene t of applying feature selection for the classi cation process is twofold. Firstly, the elimination of unnecessary features decreases the data set noise, which degrades the quality of the classi cation model. Secondly, the problem dimension is decreased, thus the e ciency is increased. Feature selection problem is very e - ciently solved by the proposed method. The classi cation quality is in the majority of cases (instances) improved relative to the methods from literature. For some of the instances, computational times are up to several hundred times smaller than those of the competing methods. Feature weighting and parameter selection problem share similar underlying solution representation, based on the vectors of real values. Since the representation of charged particles is based on the same underlying domain, the transition from the particle to the solution domain behaves smoothly. The quality of the method for iv feature weighting is demonstrated through nearest neighbors classi cation model. The testing of the method is conducted on di erent collection of instances, and after that, the comparison to several methods from literature is made. In the majority of cases, the proposed method outperformed the comparison methods. The parameter selection, in classi cation, has a great impact on the classi cation quality. The proposed method for parameter selection is applied on the support vector machihe, which has a complex parametric structure when the number of parameters and the size of their domains is in question. By using heuristic initialization procedure, the detection of high quality regions for parameter combinations is accelerated. Exhaustive tests are performed on various instances in terms of their dimension and feature structure: homogenous and heterogeneous. Single kernel learning is adopted for homogenous, and multiple kernel learning for heterogeneous instances. The comparison with methods from literature showed superiority of the proposed method when single and multiple kernel learning based on radial basis function is considered. The method shows to be competitive in other cases. All proposed methods improved the classi cation quality. Because of the way, the problem is being solved, all three methods can be generalized and applied to a wide class of classi cation models and/or classi cation problem. URI: http://hdl.handle.net/123456789/4234 Dateien zu dieser Ressource: 1
phdAleksandarKartelj.pdf ( 2.121Mb ) -
Grbić, Milana (Beograd , 2020)[more][less]
Zusammenfassung: In this dissertation some actual problems of bioinformatics and computational biology are explored,together with the methods for solving them. The following problems are considered: partitioning ofsparse biological networks intok-plexsubnetworks, prediction of the role of metabolites in metabolicreactions, partitioning of biological networks into highly connectedcomponents and the problem ofidentification of significant groups of proteins by adding new edges to the weighted protein interacti-ons network. The aforementioned problems have theoretical importance in areas of machine learningand optimization, and practical application in biological research. Inaddition to solving the afore-mentioned problems from the computational aspect, the dissertation explores further application ofthe obtained results in the fields of biology and biochemistry, as well as the integration of resultswithin existing bioinformatics tools.The problem of predicting the role of metabolites in metabolic reactions is solved by a predictivemachine learning method based on the conditional random fields, whilefor the remaining threeproblems the algorithams based on variable neighbourhood search are developed. For solving theproblem of identification of significant groups of proteins by adding new edges to the weighted proteininteractions network, the variable neighbourhood search is only the first phase of the proposedsolution, while in the second and the third phase of the proposed method, the integration withadditional biological information and bioinformatics tools are performed.The proposed computational methods of partitioning and groupingin biological networks confirmexisting findings in a new manner and lead to new discoveries about biological elements and theconnections between them. By solving these problems and by interpreting the obtained resultsin this dissertation, a scientific contribution was made to the scientific field of computer science,particularly to the scientific disciplines of bioinformatics and computational biology. URI: http://hdl.handle.net/123456789/5088 Dateien zu dieser Ressource: 1
grbic_Milana_disertacija.pdf ( 8.740Mb ) -
Jovanović, Jasmina (Beograd , 2022)[more][less]
Zusammenfassung: The analysis of biological sequence similarity between different species is significant in identifying functional, structural or evolutionary relationships among the species. Biological sequence similarity and analysis of newly discovered nucleotide and amino acid sequences are demanding tasks in bioinformatics. As biological data is growing exponentially, new and innovative algorithms are needed to be constantly developed to get faster and more effective data processing. The challenge in sequence similarity analysis algorithms is that sequence does not always have obvious features and the dimension of sequence features may be very high for applying regular feature selection methods on sequences. It is important to have a simple and effective algorithm for determining biological sequence relationships. This thesis proposes two new methods for sequence transformation in feature vectors that takes into consideration statistically significant repetitive parts of analyzed sequences, as well as includes different approaches for determination of nucleotide sequence similarity and sequence classification for predicting taxonomy groups of biological sequence data. The first method is based on information theory and fact that both position and frequency of repeated sequences are not expected to occur with the identical presence in a random sequence of the same length. The second method includes building signatures of biological sequences and profiles of taxonomic classes based on repetitive parts of sequences and distances between these repeats. Proposed methods have been validated on multiple data sets and compared with results obtained using different well known and accepted methods in this field like BLAST, Clustal Omega and methods based on k-mers. Resulted precision for proposed methods is close to values provided for existing methods for the majority of tested data-sets, and time performance depends strictly to used infrastructure and sequence type. Methods provide results that are comparable with other commonly used methods focused on resolving the same problem, taking into consideration statistically significant repetitive parts of sequences with different characteristics. URI: http://hdl.handle.net/123456789/5440 Dateien zu dieser Ressource: 1
JasminaJovanovic.pdf ( 3.984Mb ) -
Perović, Vladimir (Beograd , 2013)[more][less]
Zusammenfassung: Although long-range intermolecular interactions (interactions acting on distances >5Å) play an important role in recognition and targeting between molecules in biological systems, there is no one appropriate software package allowing use of this important property in investigation of biologically active molecules. The multifunctional EIIP/ISM software, which is based on physical parameters determining long-range molecular properties, was developed in this thesis. This novel and unique platform allows (i) investigation of protein-protein and protein-small molecule interactions, (ii) analysis of structure/function relationship of proteins, (iii) assessment of biological effects of mutations in proteins, (iv) monitoring of the functional evolution of proteins, (v) ―de novo‖ design of molecules with desired biological function and (vi) selection of candidate therapeutic molecules. Results of application of the EIIP/ISM platform on diverse problems (e.g. the evolution of influenza A viruses, assessment of biological effects of mutations on the LPL protein, representing a risk factor for cardiovascular diseases, identification of therapeutic targets for HIV and influenza viruses, virtual screening of molecular libraries for candidate antibiotics and anti-HIV drugs) which are presented in this thesis, confirm the applicability of this platform on broad spectrum of problems in molecular biology, biomedicine and pharmacology. URI: http://hdl.handle.net/123456789/4230 Dateien zu dieser Ressource: 1
phdPerovic_Vladimir.pdf ( 11.95Mb ) -
Đenić, Aleksandar (Beograd , 2018)[more][less]
Zusammenfassung: This pap er considers two discrete lo cation problems: Bus Terminal Lo cation Problem (BTLP) and Long-term Care Facility Lo cation Problem (LTCFLP). Vari- able Neighb orho o d Search (VNS) metho d for solving BTLP and LTCFLP is pre- sented in this pap er. VNS is a single-solution based metaheuristic based on system- atic change of neighb orho o ds while searching for optimal solution of the problem. It consists two main phases: shake phase and lo cal search phase. BTLP is a discrete lo cation problem which considers lo cating bus terminals in order to provide the highest p ossible quality of public service to the clients. Clients are presented as public transp ortation stations, such as bus or metro stations. VNS algorithm is used for solving BTLP. This algorithm uses improved lo cal search based on e cient neighb orho o d interchange. VNS is parallelized (PVNS) which leads to signi cant time improvement in function of the pro cessor core count. Computa- tional results show that prop osed PVNS metho d improves existing results from the literature in terms of quality. Larger instances, based on instances from the Trav- eling Salesman Problem library, are presented and computational results for those instances are rep orted. LTCFLP is created as a part of health care infrastructure planning in South Korea. Clients are considered as groups of patients with a need of long-term health care, while established facilities present lo cations where the centers that provide health care services should b e built. Prede ned are n lo cations where centers are to b e established. This problem seeks at most K lo cations to establish health centers so they are to b e equally loaded with clients demand. For solving LTCFLP, by using VNS algorithm, data structure based on fast interchange is presented. It reduces the time complexity of one iteration of lo cal search algorithm to O ( n · max( n,K 2 )) comparing to the known time complexity from the literature O ( K 2 · n 2 ) . Reduced time complexity of the presented VNS leads to b etter quality solutions, due to larger numb er of VNS iterations that can b e p erformed in less computational time. This pap er presents computational results that outp erform the b est known results from the literature. URI: http://hdl.handle.net/123456789/4744 Dateien zu dieser Ressource: 1
Aleksandar_Djenic_phd.pdf ( 2.183Mb )