Search

Razvrščanje podatkov v gruče je slabo pogojeni problem in dokazano je, da algoritem, ki bi izpolnjeval vse predpostavke dobrega razvrščanja, ne obstaja. To je glavni razlog za obstoj velikega števila algoritmov za razvrščanje, ki temeljijo na raznovrstnih teoretičnih osnovah – med njimi je tudi znan algoritem Kohonenove samo-organizirajoče mreže (SOM). Na žalost nam naučena mreža SOM ne ponudi eksplicitno izražene strukture gruč v podatkih, zato navadno posegamo uporabimo dodaten korak, na katerem združujemo posamezne enote mreže v gruče. V disertaciji predstavljamo doprinos k dvonivojskemu razvrščanju z mrežo SOM, pri čemer uporabljamo principe zakona gravitacije. Predlagan algoritem za gravitacijsko razvrščanje samo-organizirajoče mreže (gSOM) je sposoben odkriti gruče zapletene in ne zgolj hipersferične oblike. Poleg tega algoritem gSOM sam določi število gruč v podatkih. Opravili smo primerjavo z nekaterimi drugimi tehnikami razvrščanja na umetnih in realnih podatkih. Izkaže se, da gSOM doseže obetavne rezultate, še posebej na podatkih o izraženosti genov. Algoritem, ki bi znal rešiti vse probleme razvrščanja ne obstaja, zato je koristno analizirati podatke skozi večkratno razvrščanje. Pri tem nastane množica razvrstitev in tvorijo ansambel razvrstitev. Metode ansamblov za razvrščanje so se pojavile nedavno kot učinkovit pristop k stabilizaciji in izboljšanju delovanja enostavnih algoritmov za razvrščanje. Razvrščanje z ansambli je v osnovi sestavljeno iz dveh korakov: gradnja ansambla razvrstitev z enostavnimi metodami in združevanje dobljenih rešitev v sporazumno razvrstitev podatkov. Da bi olajšali korak združevanja v sporazum, je bil predlagan postopek uteževanja razvrstitev v ansamblu, ki skuša ovrednotiti pomembnost posameznih članov ansambla. Eden od načinov za analizo pomembnosti razvrstitev (PRA) je uporaba notranjih kazalcev veljavnosti razvrstitev. Na tem področju smo napravili dva prispevka: najprej predlagamo nov notranji ocenjevalni kazalec, imenovan DNs, ki razširja Dunnov kazalec in je osnovan na iskanju najkrajših poti v Gabrielovem grafu nad podatki drugi prispevek je povezan z nadgradnjo obstoječega pristopa uteženega ansambla z dodatnim korakom redukcije, ki sledi koraku ocenjevanja razvrstitev v ansamblu. Razvit postopek analize pomembnosti razvrstitev z redukcijo (PRAr) se obnese zadovoljivo, ko ga vključimo v tri funkcije za iskanje sporazumne razvrstitve, pri čemer vse funkcije temeljijo na principu kopičenja dokazov. V disertaciji se dotikamo vseh glavnih področij razvrščanja podatkov: ustvarjanje podatkov, analiza podatkov z enostavnimi algoritmi za razvrščanje, ocenjevanje razvrstitev z notranjimi in zunanjimi kazalci veljavnosti ter razvrščanje z ansambli s poudarkom na uteženih različicah. Vse predlagane doprinose smo primerjali s trenutno aktualnimi metodami na podatkih iz različnih problemskih domen. Rezultati kažejo na uporabnost predlaganih metod v kontekstu strojnega učenja. The clustering is an ill-posed problem and it has been proven that there is no algorithm that would satisfy all the assumptions about good clustering. This is why numerous clustering algorithms exist, based on various theories and approaches, one of them being the well-known Kohonen’s self-organizing map (SOM). Unfortunately, after training the SOM there is no explicitly obtained information about clusters in the underlying data, so another technique for grouping SOM units has to be applied afterwards. In the thesis, a contribution towards a two-level clustering of the SOM is presented, employing principles of Gravitational Law. The proposed algorithm for gravitational clustering of the SOM (gSOM) is capable of discovering complex cluster shapes, not only limited to the spherical ones, and is able to automatically determine the number of clusters. Experimental comparison with other clustering techniques is conducted on synthetic and real-world data. We show that gSOM achieves promising results especially on gene-expression data. As there is no clustering algorithm that can solve all the problems, it turns out as very beneficial to analyse the data using multiple partitions of them – an ensemble of partitions. Cluster-ensemble methods have emerged recently as an effective approach to stabilize and boost the performance of the single-clustering algorithms. Basically, data clustering with an ensemble involves two steps: generation of the ensemble with single-clustering methods and the combination of the obtained solutions to produce a final consensus partition of the data. To alleviate the consensus step the weighted cluster ensemble was proposed that tries to assess the relevance of ensemble members. One way to achieve this is to employ internal cluster validity indices to perform partition relevance analysis (PRA). Our contribution here is two-fold: first, we propose a novel cluster validity index DNs that extends the Dunn’s index and is based on the shortest paths between the data points considering the Gabriel graph on the data second, we propose an enhancement to the weighted cluster ensemble approach by introducing the reduction step after the assessment of the ensemble partitions is done. The developed partition relevance analysis with the reduction step (PRAr) yields promising results when plugged in the three consensus functions, based on the evidence accumulation principle. In the thesis we address all the major stages of data clustering: data generation, data analysis using single-clustering algorithms, cluster validity using internal end external indices, and finally the cluster ensemble approach with the focus on the weighted variants. All the contributions are compared to the state-of-art methods using datasets from various problem domains. Results are positive and encourage the inclusion of the proposed algorithms in the machine-learning practitioner’s toolbox.

The clustering is an ill-posed problem and it has been proven that there is no algorithm that would satisfy all the assumptions about good clustering. This is why numerous clustering algorithms exist, based on various theories and approaches, one of them being the well-known Kohonen’s self-organizing map (SOM). Unfortunately, after training the SOM there is no explicitly obtained information about clusters in the underlying data, so another technique for grouping SOM units has to be applied afterwards. In the thesis, a contribution towards a two-level clustering of the SOM is presented, employing principles of Gravitational Law. The proposed algorithm for gravitational clustering of the SOM (gSOM) is capable of discovering complex cluster shapes, not only limited to the spherical ones, and is able to automatically determine the number of clusters. Experimental comparison with other clustering techniques is conducted on synthetic and real-world data. We show that gSOM achieves promising results especially on gene-expression data. As there is no clustering algorithm that can solve all the problems, it turns out as very beneficial to analyse the data using multiple partitions of them – an ensemble of partitions. Cluster-ensemble methods have emerged recently as an effective approach to stabilize and boost the performance of the single-clustering algorithms. Basically, data clustering with an ensemble involves two steps: generation of the ensemble with single-clustering methods and the combination of the obtained solutions to produce a final consensus partition of the data. To alleviate the consensus step the weighted cluster ensemble was proposed that tries to assess the relevance of ensemble members. One way to achieve this is to employ internal cluster validity indices to perform partition relevance analysis (PRA). Our contribution here is two-fold: first, we propose a novel cluster validity index DNs that extends the Dunn’s index and is based on the shortest paths between the data points considering the Gabriel graph on the data; second, we propose an enhancement to the weighted cluster ensemble approach by introducing the reduction step after the assessment of the ensemble partitions is done. The developed partition relevance analysis with the reduction step (PRAr) yields promising results when plugged in the three consensus functions, based on the evidence accumulation principle. In the thesis we address all the major stages of data clustering: data generation, data analysis using single-clustering algorithms, cluster validity using internal end external indices, and finally the cluster ensemble approach with the focus on the weighted variants. All the contributions are compared to the state-of-art methods using datasets from various problem domains. Results are positive and encourage the inclusion of the proposed algorithms in the machine-learning practitioner’s toolbox.

Clustering or cluster analysis is a fundamental machine learning task, which is, unfortunatelly, an ill-posed problem, caused by large diversity of problem domains. Many different approaches have been used to solve it, which consequently reflects as a long list of clustering methods. Moreover, it is hard to determine, which clustering of particular data is better than another, because there does not exist an universal similarity metric, which would be the most appropriate for all different problems. In the thesis, four chosen methods for clustering are being examined, each of which has its interesting features. These are: KMC, ECMC, EM GMM in CSC. In addition, new criteria for the evaluation of clustering correctness appear, which are inherently subject to a peer comparison. My intention was to carry out a comprehensive analysis of the chosen methods and objectively evaluate the results of the clustering of individual typical problem domain. To achieve this, four internal and six external evaluation criteria or indices were used. On their basis final evaluation of the effectiveness of various methods is given. Several synthetic and real data sets on which the clustering has been performed out have been selected to reflect the typical problems in this field. The final results of the comparison shows that the application of knowledge of information theory, which exploits novel CSC method, contribute to a better outcome depending on the selected criteria and the data sets. It also opens up considerable potential to continue its improvement and is also the motivation for using alternative approaches to solve the clustering problem.

CmDock je odprtokodni program za simulacijo molekulskega sidranja. S pomočjo genetskega algoritma išče optimalno konformacijo manjše molekule, vezane na površino beljakovine. Program smo razširili z dvema izvedbama nove cenilne funkcije na osnovi odsekoma linearnega potenciala. Prva se izvaja na enem jedru centralne procesorske enote, druga pa izkorišča zmožnost vzporednega računanja na grafičnih pospeševalnikih in gradi na prototipu v ogrodju OpenCL. Rezultate nove cenilne funkcije smo primerjali z rezultati obstoječe, pri čemer smo ocenjevali točnost ter hitrost izračuna na treh izbranih testnih kompleksih. Pri enem izmed njih je rezultat bolj točen, pri ostalih dveh pa zaostajamo za obstoječo cenilno funkcijo. Izvajanje programa je z novo cenilno funkcijo na grafičnem pospeševalniku 3-krat hitrejše kot na centralni procesorski enoti ter 8-krat hitrejše kot z obstoječo cenilno funkcijo. Del, ki se v celoti izvede vzporedno, porabi na grafičnem pospeševalniku do 76-krat manj časa. CmDock is an open-source program intended for simulating molecular docking. It leverages a genetic algorithm to search for the optimal conformation of a small molecule (ligand) docked onto the surface of a protein. This thesis presents an improvement to the mentioned program by implementing two versions of a scoring function based on piecewise linear potential. The first one executes on the central processing unit (CPU), while the second one utilizes parallel computation capabilities of graphics processing units (GPUs) and is based on a prototype in OpenCL. We have compared the results of both versions we evaluated the accuracy and speed on three protein-ligand complexes. Compared to the established scoring function, we achieved better accuracy when docking onto one of the three proteins, while docking on the other two was less successful. CmDock utilizing GPU accelerated scoring function achieves around 3x speedup over the equivalent CPU-based version and is around 8x faster than the existing version. Inspecting only the fully parallelizable part, we have observed up to 76x reduction in computation time when using GPU acceleration.

Svet interneta stvari postaja vse bolj priljubljen, saj nam olajša vsakodnevna opravila. Pri tem lahko nastanejo velike množice podatkov in z njimi potreba po njihovi obdelavi in analizi, kjer nam lahko področje strojnega učenja dodatno razširi funkcionalnost sistemov. Hkrati pa tovrstna tematika vse pogosteje postaja del industrijskih obratov, kjer sta hitrost in kakovost obdelave primarna vidika. Namen izbrane diplomske naloge je meritev in analiza električnega toka električnega mešalnika za hrano ter posledično napoved vsebine naprave s pomočjo metod strojnega učenja. Električni mešalnik je pri tem zgolj poceni in praktičen nadomestek pravih industrijskih naprav. Izvedli smo več meritev posameznih vhodnih sestavin mešalnika, ki smo jih kasneje uporabili v štirih metodah strojnega učenja. Metodam smo nato ovrednotili čas izvajanja in natančnost napovedovanja. Pogosto smo uspeli napovedati stanje naprave, a smo opazili največ težav pri napovedovanju vhodnih sestavin, ki so bile po gostoti oziroma viskoznosti najbolj podobne. The world of IoT is becoming increasingly popular as it makes everyday tasks easier but can lead to large data sets. We need to process and analyze the obtained data, and this is where the field of machine learning can further extend the functionality of our systems. At the same time, IoT and machine learning are becoming part of industrial plants, where speed and quality of processing are primary considerations. The aim of this diploma thesis is to measure and analyze the electric current of an electric food mixer and predict the content of the device using machine learning algorithms. An electric mixer is merely a cheap and practical substitute for real industrial devices. We conducted several measurements of individual input ingredients of the mixer, which were subsequently used in four machine learning methods. The methods were then evaluated for their execution time and prediction accuracy. While we were often able to accurately predict the state of the device, we observed the greatest difficulty in predicting input ingredients that were most similar in density or viscosity.

Z razvojem sodobnega grafičnega sklada Linux se je pred približno desetletjem pojavil protokol Wayland. Ta z arhitekturo, skladno s sodobnimi potrebami, na glavno mesto postavlja skladatelja. Zaradi relativne novosti protokola Wayland ta le počasi zamenjuje uveljavljeni protokol X in v različnih implementacijah pozna določene ovire za uporabnost v obliki manjkajočih funkcionalnosti. V diplomski nalogi smo razvili novega skladatelja Wayland, ki temelji na knjižnici wlroots in ponuja celovitejšo uporabniško izkušnjo. Razvita rešitev ima vgrajeno podporo za deljenje in zajem zaslona, napredne poteze s sledilnimi ploščicami in polno podporo za odložišče. S temi vgrajenimi funkcionalnostmi prispeva k prepoznavi in razširjenosti Waylanda kot polno funkcionalne zamenjave za X. With the development of a modern Linux graphical stack, the Wayland protocol emerged about a decade ago. Its architecture better accomodates the demands of modern hardware, software and security, and brings the compositor to the forefront. The relative novelty of the Wayland protocol has caused its adoption and replacement of the old X protocol to be somewhat slow, and has left some implementations with varying obstacles for usability in terms of missing modern desktop features. In this thesis we have developed a novel Wayland compositor, based on the wlroots library, that solves some of the usability obstacles we have encountered. The developed solution includes built-in support for screencasting and screensharing, advanced touchpad gesture support and full clipboard support. With these features it contributes to the recognition and adoption of Wayland as a fully-featured replacement for X.

Kot del relativno novega področja je komunikacija z napravami IoT še vedno odprta tema. Glavna pomanjkljivost dosedanjih rešitev je fragmentacija in kompleksnost uporabe. Za projekte manjšega obsega zato postane problem odkrivanja naprav in komunikacije med napravami precej velika ovira. Da bi jo presegli, smo razvili protokol MultiCom in množico knjižnic, ki omogočajo celovito in preprosto rešitev za komunikacijo mobilne naprave z napravami IoT. Rešitev smo preizkusili tudi z vključitvijo v obstoječi projekt izdelave mobilne aplikacije za upravljanje pametnih infrardečih grelcev. As part of a relatively new field, the problem of communication with IoT devices remains an open question. The main disadvantage of current solutions is the fragmentation and complex usage. For smaller projects, the problem of device discovery and communication between devices becomes a relatively big burden. To overcome it, we have developed the protocol MultiCom and a set of libraries, enabling a complete and simple solution for a mobile device to communicate with IoT devices. We have set ourselves a goal of solving this problem with a protocol and a set of libraries that simplify the communication of a mobile device with an IoT device. We tested the solution by including it in an existing project of developing an application to manage smart infrared heaters.

V diplomskem delu smo razvili vzporedni genetski algoritem, ki se izvaja na heterogenih računalniških arhitekturah za potrebe simulacije molekulske dinamike. Razviti algoritem uporablja empirično cenilno funkcijo za ocenjevanje rešitev. Simulirali smo sidranje molekul v receptorsko mesto proteina in iskali optimalen položaj molekule. Uporabili smo razvojno ogrodje OpenCL. Analizirali smo konvergenco in učinkovitost algoritma. Uporabljeno merilo učinkovitosti je bil izvajalni čas simulacije. Za testne primere smo uporabili dva liganda. Algoritem smo preizkusili in ovrednotili na dveh grafičnih pospeševalnikih in večjedrnem procesorju. Vzporedni algoritem konvergira in vrača pričakovane rezultate. Za učinkovitejšo rabo grafične procesne enote in večje pohitritve je potrebno algoritem dodatno optimizirati. In this thesis we developed a parallel genetic algorithm which can run on heterogeneous systems to simulate molecular dynamics. The algorithm uses an empirical scoring function. We simulated molecule docking to a receptor protein and searched for optimal molecule position. We used OpenCL framework. We analysed the convergence and efficiency of the algorithm. We were primarily concerned with the simulation execution time. We used two ligands as test cases. The algorithm was evaluated on two graphics accelerators and a multi-core processor. Parallel algorithm converges and returns the expected results. For more efficient use of a graphics processing unit and achieving better speedup the algorithm needs further optimization.

Diplomsko delo opisuje potek integracije avtomatiziranega funkcionalnega testiranja spletne trgovine. Glede na predstavljene zahteve naročnika je bilo uporabljeno orodje Katalon. Začnemo pri prvih korakih uporabe, kjer smo se razvijalci spoznavali z delovanjem in zmožnostmi orodja. Nadaljujemo z reševanjem odkritih problemov in optimizacijami za lažji razvoj ter vzdrževanje testnih primerov. Zaključimo z integracijo Katalon TestOps, ki služi kot orodje za agregacijo rezultatov testiranj in umestitvijo pogona avtomatiziranih testnih skriptov v sistem neprekinjene integracije in dostave. This thesis describes the integration of an automated functional testing tool for an online store. According to the requirements of the client, the Katalon tool was used. First, we present how the developers got to know the functionalities and capabilities of the tool. Next, we describe our approaches to solve the identified problems and implement optimizations to facilitate the development and maintenance of test cases. Finally, we conclude with the integration of Katalon TestOps, which serves as a tool for aggregating test results and executing automated test scripts in continuous integration and delivery system.