Vykstant intensyviai konkurencijai prekybos įmonės, norėdamos išlikti rinkoje ir gauti pelną, jungiasi į klasterius, nes suformavus prekybos įmonių junginius, jų vaidmuo ir galia rinkoje labai išauga. Tai pagal geografinį principą sukoncentruotos tarpusavyje sąveikaujančios, vykdančios bendrą veiklą ir savo specifine veikla papildančios viena kitą bendrovės, specializuoti tiekėjai, paslaugų teikėjai ir įvairios organizacijos (pvz., universitetai, standartizacijos agentūros, prekybiniai susivienijimai). Darbo objektas – maisto pramonės šakos grūdų perdirbimo sektoriaus įmonės. Darbo tikslas – nustatyti maisto pramonės šakos grūdų perdirbimo sektoriaus klasterizacijos įtaką šalies konkurencingumui. Darbo tikslui pasiekti ir uždaviniams pasiekti naudojami šiais tyrimo metodais: mokslinės literatūros analize, lyginamąja analize, žvalgomuoju tyrimu, statistinių duomenų ir dokumentų analize, interviu, įmonės dokumentų analize. Darbas susideda iš trijų skyrių. Pirmajame skyriuje nagrinėjamas pramonės konkurencingumas ir jį lemiantys veiksniai, atliekant konkurencingumo sampratos analizę bei tiriant konkurencingumą sąlygojančius veiksnius. Antrajame skyriuje analizuojamas klasteris, kaip stiprus postūmis šalies konkurencingumui, atliekant klasterio sampratos, bruožų ir tipų analizę bei įvertinant klasterių įtaką konkurencingumui. Trečiajame skyriuje maisto pramonės šakos grūdų perdirbimo sektoriaus klasterio įtaka šalies konkurencingumui. Darbo pabaigoje pateikiamos reikšmingos... [toliau žr. visą tekstą] / Under an intensive competition, trading enterprises unite into clusters in order to remain in market to get profit. Clusters are geographic concentrations of interconnected companies, specialized suppliers, service providers, firms in related industries, and associated institutions in a particular field that compete but also cooperate. The aim of the work is to ascertain the influence of clusters of food industry on competitiveness in the country. The object of the work is food industry enterprises. The methods of research are the analysis of science literature, comparative analysis, reconnaissance research, the analysis of statistic data and documents, interview, the analysis of the company documents. The research was carried out in September 2004 and it lasted till the May of 2006. The work consists of three parts. The first part defines the concept of competitiveness and explores the factors which cause competition. Clusters as a strong stimulus to country competition is analyzed in the second part. The third part carries out a research how food industrial enterprise clusters influences the country competitiveness. At the end of work the significant conclusion is made. Clusters have an effect on competitiveness in these ways: increase the productivity of companies forming a cluster, encourage innovations and hasten their rate of origin, encourage the origin of new trades. And in this expands the limits and influence of a cluster. Grain and its foodstuffs production cluster... [to full text]

This work is devoted to the speech signal clustering analysis problem. The main methods of cluster analysis were reviewed in this work and clusterization algorithm based on vector quantization was created. The speaker identification experiments were performed in which dependence of identification accuracy and computational complexity from number of clusters was investigated.

Uncompressed multimedia data requires considerable storage capacity and transmission bandwidth. Despite rapid progress in mass-storage density, processor speeds, and digital communication system performance, demand for data storage capacity and data-transmission bandwidth continues to outstrip the capabilities of available technologies. The recent growth of data intensive multimedia-based web applications even more sustained the need for more efficient ways to encode such data. There are two types of image compression schemes – lossless and lossy algorithms. In lossless compression schemes, the reconstructed image, after compression, is numerically identical to the original image. However lossless compression can only achieve a modest amount of compression. An image reconstructed following lossy compression contains degradation relative to the original. Often this is because the compression scheme completely discards redundant information. However, lossy schemes are capable of achieving much higher compression. The aim of this research is to create an efficient lossy image compression algorithm, using heuristic data clusterization methods; perform experiments of the new algorithm, measure its performance, analyze advantages and disadvantages of the proposed method, propose possible improvements and compare it with other popular algorithms. In this paper is presented new algorithm for image compression, which uses data base of popular image fragments. Proposed algorithm is... [to full text]

In the context of adaptive intellectual learning environment (VLE) possibility of using conceptual clustering algorithms is analyzed, trying to accomplish the ability of software agent to "feel" the changing environment and recognise the states it is in. Agent environment is understood as interface between the user of VLE and the students model, which is stored in the VLE and is constantly changing. Agents ability to "feel" is understood as agents ability to classify students, based on their knowledge level, which changes in the learning process. Using conceptual clustering algorithms found in the literature, we are trying to choose one which is most suited for the problem area, modifying it to model real data.

<p>Disertacija se se bavi analizom mogućnosti brze, efikasne<br />i pouzdane klasterizacije masivnog skupa neuralnih<br />snimaka na osnovu probabilističkih parametara procenjenih<br />iz obrazaca generisanja akcionih potencijala, tzv.<br />&quot;spajkova&quot;, na izlazu pojedinih neurona. Neuralna<br />aktivnost se grubo može podeliti na periode intezivne,<br />umerene i niske aktivnosti. Shodno tome, predložena je<br />gruba dekompozicija neuralne aktivnosti na tri moda koja<br />odgovaraju navedenim obrascima neuralne aktivnosti, na<br />osnovu dobro poznatog Gilbert-Eliot modela. Modovi su<br />dodatno ra&scaron;članjeni na sopstvena stanja na osnovu osobina sukcesivnih spajkova, omogućujući finiji, kompozitni<br />opis neuralne aktivnosti. Za svaki neuron empirijski se<br />procenjuju probabilistički parametri grube dekompozicije<br />- na osnovu Gilbert-Eliotovog modela i finije dekompozicije<br />- na osnovu sopstvenih stanja modova, obezbeđujući<br />željeni skup deskriptora. Dobijeni deskriptori<br />koriste se kao obeležja nekoliko algoritama klasterizacije<br />nad simuliranim i eksperimentalnim podacima. Za generisanje<br />simuliranih podataka primenjen je jednostavan<br />model za generisanje akcionih potencijala različitih<br />oscilatornih pona&scaron;anja pobuđujućih i blokirajućih kortikalnih<br />neurona. Validacija primene probabilističkih parametara<br />za klasterizaciju rada neurona izvr&scaron;ena je na<br />osnovu estimacije parametera nad generisanim neuralnim<br />odzivima. Eksperimentalni podaci su dobijeni<br />snimanjem kortikografskih signala iz dorzalnog anteriornog<br />cingularanog korteksa i lateralnog prefrontalnog<br />korteksa korteksa budnih rezus majmuna. U okviru predloženog<br />protokola evaluacije različitih pristupa<br />klasterizacije testirano je nekoliko metoda. Klasterizacija<br />zasnovana na akumulaciji dokaza iz ansambla particija<br />dobijenih k-means klasterovanjem dala je najstabilnije<br />grupisanje neuralnih jedinica uz brzu i efikasnu implementaciju.<br />Predložena empirijska karakterizacija može da<br />posluži za identifikaciju korelacije sa spolja&scaron;njim stimulusima,<br />akcijama i pona&scaron;anjem životinja u okviru<br />eksperimentalne procedure. Prednosti ovog postupka za<br />opis neuralne aktivnosti su brza estimacija i mali skup<br />deskriptora. Računarska efikasnost omogućuje primenu<br />nad obimnim, paralelno snimanim neuralnim podacima u<br />toku snimanja ili u periodima od interesa za identifikaciju<br />aktiviranih i povezanih zona pri određenim aktivnostima.</p> / <p>The advances in extracellular neural recording techniques<br />result in big data volumes that necessitate fast,<br />reliable, and automatic identification of statistically<br />similar units. This study proposes a single framework<br />yielding a compact set of probabilistic descriptors that<br />characterise the firing patterns of a single unit. Probabilistic<br />features are estimated from an inter-spikeinterval<br />time series, without assumptions about the firing distribution or the stationarity. The first level of proposed<br />firing patterns decomposition divides the inter-spike<br />intervals into bursting, moderate and idle firing modes,<br />yielding a coarse feature set. The second level identifies<br />the successive bursting spikes, or the spiking acceleration/<br />deceleration in the moderate firing mode, yielding<br />a refined feature set. The features are estimated from<br />simulated data and from experimental recordings from<br />the lateral prefrontal cortex in awake, behaving rhesus<br />monkeys. An effcient and stable partitioning of neural<br />units is provided by the ensemble evidence accumulation<br />clustering. The possibility of selecting the number of<br />clusters and choosing among coarse and refined feature<br />sets provides an opportunity to explore and compare<br />different data partitions. The estimation of features, if<br />applied to a single unit, can serve as a tool for the firing<br />analysis, observing either overall spiking activity or the<br />periods of interest in trial-to-trial recordings. If applied to<br />massively parallel recordings, it additionally serves as an<br />input to the clustering procedure, with the potential to<br />compare the functional properties of various brain<br />structures and to link the types of neural cells to the<br />particular behavioural states.</p>

<p>In the current information age, massive amounts of data are gathered, at a rate prohibiting their effective structuring, analysis, and conversion into useful knowledge. This information overload is manifested both in large numbers of data objects recorded in data sets, and large numbers of attributes, also known as high dimensionality. This dis-sertation deals with problems originating from high dimensionality of data representation, referred to as the &ldquo;curse of dimensionality,&rdquo; in the context of machine learning, data mining, and information retrieval. The described research follows two angles: studying the behavior of (dis)similarity metrics with increasing dimensionality, and exploring feature-selection methods, primarily with regard to document representation schemes for text classification. The main results of the dissertation, relevant to the first research angle, include theoretical insights into the concentration behavior of cosine similarity, and a detailed analysis of the phenomenon of hubness, which refers to the tendency of some points in a data set to become hubs by being in-cluded in unexpectedly many <em>k</em>-nearest neighbor lists of other points. The mechanisms behind the phenomenon are studied in detail, both from a theoretical and empirical perspective, linking hubness with the (intrinsic) dimensionality of data, describing its interaction with the cluster structure of data and the information provided by class la-bels, and demonstrating the interplay of the phenomenon and well known algorithms for classification, semi-supervised learning, clustering, and outlier detection, with special consideration being given to time-series classification and information retrieval. Results pertaining to the second research angle include quantification of the interaction between various transformations of high-dimensional document representations, and feature selection, in the context of text classification.</p> / <p>U tekućem &bdquo;informatičkom dobu&ldquo;, masivne količine podataka se<br />sakupljaju brzinom koja ne dozvoljava njihovo efektivno strukturiranje,<br />analizu, i pretvaranje u korisno znanje. Ovo zasićenje informacijama<br />se manifestuje kako kroz veliki broj objekata uključenih<br />u skupove podataka, tako i kroz veliki broj atributa, takođe poznat<br />kao velika dimenzionalnost. Disertacija se bavi problemima koji<br />proizilaze iz velike dimenzionalnosti reprezentacije podataka, često<br />nazivanim &bdquo;prokletstvom dimenzionalnosti&ldquo;, u kontekstu ma&scaron;inskog<br />učenja, data mining-a i information retrieval-a. Opisana istraživanja<br />prate dva pravca: izučavanje pona&scaron;anja metrika (ne)sličnosti u odnosu<br />na rastuću dimenzionalnost, i proučavanje metoda odabira atributa,<br />prvenstveno u interakciji sa tehnikama reprezentacije dokumenata za<br />klasifikaciju teksta. Centralni rezultati disertacije, relevantni za prvi<br />pravac istraživanja, uključuju teorijske uvide u fenomen koncentracije<br />kosinusne mere sličnosti, i detaljnu analizu fenomena habovitosti koji<br />se odnosi na tendenciju nekih tačaka u skupu podataka da postanu<br />habovi tako &scaron;to bivaju uvr&scaron;tene u neočekivano mnogo lista k najbližih<br />suseda ostalih tačaka. Mehanizmi koji pokreću fenomen detaljno su<br />proučeni, kako iz teorijske tako i iz empirijske perspektive. Habovitost<br />je povezana sa (latentnom) dimenzionalno&scaron;ću podataka, opisana<br />je njena interakcija sa strukturom klastera u podacima i informacijama<br />koje pružaju oznake klasa, i demonstriran je njen efekat na<br />poznate algoritme za klasifikaciju, semi-supervizirano učenje, klastering<br />i detekciju outlier-a, sa posebnim osvrtom na klasifikaciju vremenskih<br />serija i information retrieval. Rezultati koji se odnose na<br />drugi pravac istraživanja uključuju kvantifikaciju interakcije između<br />različitih transformacija vi&scaron;edimenzionalnih reprezentacija dokumenata<br />i odabira atributa, u kontekstu klasifikacije teksta.</p>