This paper is focused on object learning and recognizing in the image and in the image stream. More specifically on learning and recognizing humans or theirs parts in case they are partly occluded, with possible usage on robotic platforms. This task is based on features called Histogram of Oriented Gradients (HOG) which can work quite well with different poses the human can be in. The human is split into several parts and those parts are detected individually. Then a system of voting is introduced in which detected parts votes for the final positions of found people. For training the detector a linear SVM is used. Then the Kalman filter is used for stabilization of the detector in case of detecting from image stream.

This thesis is focused on detecting multiple sclerosis lesions from magnetic resonance images. Correctly retrieved lesions are very important for medical diagnosis. Detection of lesions using machine learning techniques is quite challenging because of large variability in size, shape and position of lesions in the brain. In the practical part is designed base software, which after completion will classify pixels, so that is possible to find lesions of multiple sclerosis. For classification will be used Support vector machine. Theoretical part describes multiple sclerosis, basic operations performed with biomedical images and data classification.

This work explores machine learning as a tool for medical images' classification. A literary research is contained concerning both classical and modern approaches to image segmentation. The main purpose of this work is to design and implement an extension for the 3D Slicer platform. The extension uses machine learning to classify images using set parameters. The extension is tested on tomographic images obtained by nuclear magnetic resonance and observes the accuracy of the classification and usability in practice.

For two decades, researchers and clinicians have been using the diagnostic criteria for FTD to generally diagnose a patient as suffering from PPA and the criteria of Neary et al. (1998) to further specify the diagnosis as progressive nonfluent aphasia or semantic dementia. However, there were a number of PPA cases that could not be classified according to the criteria of Neary and colleagues, which led to a revision of the diagnostic clinical and research criteria for PPA by Gorno-Tempini et al. (2011). The revised criteria encompass three PPA variants (svPPA, nfvPPA, and lvPPA) with three stages characterized by increasing evidence: clinical diagnosis, imaging-supported diagnosis, and diagnosis with definite pathology. As compared to the previous diagnostic criteria, more emphasis is placed on imaging markers as supportive features. These imaging criteria were however proposed based on a purely qualitative evaluation of the literature and have not been validated so far. The aim of this thesis was to quantitatively evaluate the validity of the new diagnostic imaging criteria for PPA variants using anatomical likelihood meta-analyses (study 1) and to investigate the usefulness of these imaging criteria for the individual diagnosis of PPA patients in clinical routine using support vector machine classification (study 2).

Diese Arbeit zeigt ausgehend von einer Darstellung der theoretischen Grundlagen automatischer Textklassifikation, dass die aus der Statistical Learning Theory stammenden Support Vector Machines geeignet sind, zu einer präziseren Erkennung unerwünschter E-Mail-Werbung beizutragen. In einer Testumgebung mit einem Corpus von 20 000 E-Mails wurden Testläufe verschiedene Parameter der Vorverarbeitung und der Support Vector Machine automatisch evaluiert und grafisch visualisiert. Aufbauend darauf wird eine Erweiterung für die Open-Source-Software SpamAssassin beschrieben, die die vorhandenen Klassifikationsmechanismen um eine Klassifikation per Support Vector Machine erweitert.

There was an increase in a number of applications for a master degree program with the growth in time. It takes huge time to process all the application documents of each and every applicant manually and requires a high volume of the workforce. This can be reduced if automation is used for this process. In any case, before that, an analysis of the complete strides required in preparing was precisely the automation must be utilized to diminish the time and workforces must be finished. The application process for the applicant is actually participating in several steps. First, the applicant sends the complete scanned documents to the uni-assist; from there the applications are received by the student assistant team at the particular university to which the applicant had applied, and then they are sent to the individual departments. At the individual sections, the individual applications will be handled by leading an intensive study to know whether the applicant by their past capabilities scopes to satisfy the prerequisites of further study system to which they have applied. What's more, by considering the required points of interest of the applicant without investigating every single report, and to pack the information and diminish the preparing time for the specific division, by this postulation extend a solitary web apparatus is being produced that can procedure the application which is much dependable in the basic leadership procedure of application.

This project will explore machine learning approaches that are used in creditscoring. In this study we consider consumer credit scoring instead of corporatecredit scoring and our focus is on methods that are currently used in practiceby banks such as logistic regression and decision trees and also compare theirperformance against machine learning approaches such as support vector machines (SVM), neural networks and random forests. In our models we addressimportant issues such as dataset imbalance, model overfitting and calibrationof model probabilities. The six machine learning methods we study are support vector machine, logistic regression, k-nearest neighbour, artificial neuralnetworks, decision trees and random forests. We implement these models inpython and analyse their performance on credit dataset with 30000 observations from Taiwan, extracted from the University of California Irvine (UCI)machine learning repository.

A child dies of pneumonia every 39 seconds, and the process of preventing deaths caused by pneumonia has been considerably slower compared to other infectious diseases. Meanwhile, the traditional method of manually diagnosing patients has reached its ceiling on performance. With the support of a machine learning classification algorithm to help with the screening of pneumonia from x-ray images combined with the expertise of a physician, the identification and diagnosis of pediatric pneumonia should be both quicker and more accurate. In this study, four different types of supervised machine learning algorithms have been trained, tested, and evaluated to see which model could predict most accurately whether a patient in an x-ray image has pneumonia or not. The four models included in this study have been trained by four different supervised machine learning algorithms: logistic regression, k-nearest-neighbor, support vector machine, and neural network. The results show that KNN has the highest sensitivity, NN adapts to new data the best by not being under- or overfit. SVM had the highest balanced accuracy on both train and test data but a proportionally high difference between the in- and out-sample error. In conclusion, relatively high performance can be achieved when classifying x-ray images of pneumonia even with limited resources.

Reinforcement learning has proven itself very useful in certain areas, such as games. However, the approach has been seen as quite limited. Reinforcement-based learning has for instance not been commonly used for classification tasks as it is receiving feedback on how well it did for an action performed on a specific input. This slows the performance convergence rate as compared to other classification approaches which has the input and the corresponding output to train on. Nevertheless, this thesis aims to investigate whether reinforcement-based learning could successfully be employed on a classification task. Moreover, as sensor fusion is an expanding field which can for instance assist autonomous vehicles in understanding its surroundings, it is also interesting to see how sensor fusion, i.e., fusion between lidar and RGB images, could increase the performance in a classification task. In this thesis, a reinforcement-based learning approach is compared to a semi-classical approach. As an example of a reinforcement learning model, a deep Q-learning network was chosen, and a support vector machine classifier built on top of a deep neural network, was chosen as an example of a semi-classical model. In this work, these frameworks are compared with and without sensor fusion to see whether fusion improves their performance. Experiments show that the evaluated reinforcement-based learning approach underperforms in terms of metrics but mainly due to its slow learning process, in comparison to the semi-classical approach. However, on the other hand using reinforcement-based learning to carry out a classification task could still in some cases be advantageous, as it still performs fairly well in terms of the metrics presented in this work, e.g. F1-score, or for instance imbalanced datasets. As for the impact of sensor fusion, a notable improvement can be seen, e.g. when training the deep Q-learning model for 50 episodes, the F1-score increased with 0.1329; especially, when taking into account that the most of the lidar data used in the fusion is lost since this work projects the 3D lidar data onto the same 2D plane as the RGB images.

The significant increase of astronomical data necessitates new strategies and developments to analyse a large amount of information, which no longer is efficient if done by hand. Supervised machine learning is an example of one such modern strategy. In this work, we apply the classification technique on Gaia+2MASS+WISE data to explore the usage of supervised machine learning on large astronomical archives. The idea is to create an algorithm that recognises entries with unusual infrared properties which could be interesting for follow-up observations. The programming is executed in MATLAB and the training of the algorithms in the classification learner application of MATLAB. Each catalogue; Gaia+2MASS+WISE contains ~109, 5×108 and 7×108 (The European Space Agency 2019, Skrutskie et al. 2006, R. M. Cutri IPAC/Caltech) entries respectively. The algorithms searches through a sample from these archives consisting of 765266 entries, corresponding to objects within a <500 pc range. The project resulted in a list of 57 entries with unusual infrared properties, out of which 8 targets showed none of the four common features that provide a natural physical explanation to the unconventional energy distribution. After more comprehensive studies of the aforementioned targets, we deem it necessary for further studies and observations on 2 out of the 8 targets (Nr.1 and Nr.8 in table 3) to establish their true nature. The results demonstrate the applicability of machine learning in astronomy as well as suggesting a sample of intriguing targets for further studies. / Inom astronomi samlas stora mängder data in kontinuerligt och dess tillväxt ökar snabbt för varje år. Detta medför att manuella analyser av datan blir mindre och mindre lönsama och kräver istället nya strategier och metoder där stora datamängder snabbare kan analyseras. Ett exempel på en sådan strategi är vägledd maskininlärning. I detta arbete utnyttjar vi en vägled maskininlärnings teknik kallad klassificering. Vi använder klassificerings tekniken på data från de tre stora astronomiska katalogerna Gaia+2MASS+WISE för att undersöka användningen av denna teknik på just stora astronomiska arkiv. Idén är att skapa en algorithm som identifierar objekt med okontroversiella infraröda egenskaper som kan vara intressanta för vidare observationer och analyser. Dessa ovanliga objekt är förväntade att ha en lägre emission i det optiska våglängdsområdet och en högre emission i det infraröda än vad vanligtvis är observerad för en stjärna. Programmeringen sker i MATLAB och träningsprocessen av algoritmerna i MATLABs applikation classification learner. Algoritmerna söker igenom en samling data bestående av 765266 objekt, från katalogerna Gaia+2MASS+WISE. Dessa kataloger innehåller totalt ~109, 5×108 och 7×108 (The European Space Agency 2019, Skrutskie et al. 2006, R. M. Cutri IPAC/Caltech) objekt vardera. Det begränsade dataset som algoritmerna söker igenom motsvarar objekt inom en radie av <500 pc. Många av de objekt som algoritmerna identifierade som ”ovanliga” tycks i själva verket vara nebulösa objekt. Den naturliga förklaringen för dess infraröda överskott är det omslutande stoft som ger upphov till värmestrålning i det infraröda. För att eliminera denna typ av objekt och fokusera sökningen på mer okonventionella objekt gjordes modifieringar av programmen. En av de huvudsakliga ändringarna var att introducera en tredje klass bestående av stjärnor inneslutna av stoft som vi kallar "YSO"-klassen. Ytterligare en ändring som medförde förbättrade resultat var att introducera koordninaterna i träningen samt vid den slutgiltiga klassificeringen och på så vis, identifiering av intressanta kandidater. Dessa justeringar resulterade i en minskad andelen nebulösa objekt i klassen av ”ovanliga” objekt som algoritmerna identifierade. Projektet resulterade i en lista av 57 objekt med ovanliga infraröda egenskaper. 8 av dessa objekt påvisade ingen av det fyra vanligt förekommande egenskaperna som kan ge en naturlig förklaring på dess överflöd av infraröd strålning. Dessa egenskaper är; nebulös omgivning eller påvisad stoft, variabilitet, Hα emission eller maser strålning. Efter vidare undersökning av de 8 tidigare nämnda objekt anser vi att 2 av dessa behöver vidare observationer och analys för att kunna fastslå dess sanna natur (Nr.1 och Nr.8 i tabell 3). Den infraröda strålningen är alltså inte enkelt förklarad för dessa 2 objekt. Resultaten av intressanta objekt samt övriga resultat från maskininlärningen, visar på att klassificeringstekniken inom maskininlärning är användbart på stora astronomiska datamängder.