DNA microarrays have been used for the purpose of monitoring expression levels of thousands of genes simultaneously and identifying those genes that are differentially expressed. One of the major goals of microarray data analysis is the detection of differentially expressed genes across two kinds of tissue samples or samples obtained under two experimental conditions. A large number of gene detection methods have been developed and most of them are based on statistical analysis. However the statistical analysis methods have the limitations due to the small sample size and unknown distribution and error structure of microarray data. In this thesis, a study of ranking-based gene selection methods which have weak assumption about the data was done. Three approaches are proposed to integrate the individual ranks to select differentially expressed genes in microarray data. The experiments are implemented on the simulated and biological microarray data, and the results show that ranking-based methods outperform the t-test and SAM in selecting differentially expressed genes, especially when the sample size is small.

Software projects go through a phase of maintenance and, in case of open source projects, through hard development process. Both of these phases are prone to regressions, meaning previously working parts of system do not work anymore. To avoid this behavior, systems are being tested with long test suites, which can be sometimes time consuming. For this reason, prediction models are developed to predict software regressions using historical testing data and code changes, to detect changes that can most likely cause regression and focus testing on such parts of code. But, these predictors rely on static code analysis without deeper semantic understanding of the code. Purpose of this master thesis is to create predictor, that relies not only on static code analysis, but provides decisions based on code semantics as well.

Yes / We compare the performance of logistic regression with several alternative machine learning methods to estimate the risk of death for patients following an emergency admission to hospital based on the patients’ first blood test results and physiological measurements using an external validation approach. We trained and tested each model using data from one hospital (n=24696) and compared the performance of these models in data from another hospital (n=13477). We used two performance measures – the calibration slope and area under the curve (AUC). The logistic model performed reasonably well – calibration slope 0.90, AUC 0.847 compared to the other machine learning methods. Given the complexity of choosing tuning parameters of these methods, the performance of logistic regression with transformations for in-hospital mortality prediction was competitive with the best performing alternative machine learning methods with no evidence of overfitting. / Health Foundation; National Institute for Health Research (NIHR) Yorkshire and Humberside Patient Safety Translational Research Centre (NIHR YHPSTRC)

Pervasion of mobile devices led to a growing interest in touch-based interactions. However, multi-touch input is still restricted to direct manipulations. In current applications, gestural commands - if used at all - are only exploiting single-touch. The underlying motive for the work at hand is the conviction that a realization of advanced interaction techniques requires handy tools for supporting their interpretation. Barriers for own implementations of procedures are dismantled by providing proof of concept regarding manifold interactions, therefore, making benefits calculable to developers. Within this thesis, a recognition routine for planar, symbolic gestures is developed that can be trained by specifications of templates and does not imply restrictions to the versatility of input. To provide a flexible tool, the interpretation of a gesture is independent of its natural variances, i.e., translation, scale, rotation, and speed. Additionally, the essential number of specified templates per class is required to be small and classifications are subject to real-time criteria common in the context of typical user interactions. The gesture recognizer is based on the integration of a nearest neighbor approach into a Bayesian classification method. Gestures are split into meaningful, elementary tokens to retrieve a set of local features that are merged by a sensor fusion process to form a global maximum-likelihood representation. Flexibility and high accuracy of the approach is empirically proven in thorough tests. Retaining all requirements, the method is extended to support the prediction of partially entered gestures. Besides more efficient input, the possible specification of direct manipulation interactions by templates is beneficial. Suitability for practical use of all provided concepts is demonstrated on the basis of two applications developed for this purpose and providing versatile options of multi-finger input. In addition to a trainable recognizer for domain-independent sketches, a multi-touch text input system is created and tested with users. It is established that multi-touch input is utilized in sketching if it is available as an alternative. Furthermore, a constructed multi-touch gesture alphabet allows for more efficient text input in comparison to its single-touch pendant. The concepts presented in this work can be of equal benefit to UI designers, usability experts, and developers of feedforward-mechanisms for dynamic training methods of gestural interactions. Likewise, a decomposition of input into tokens and its interpretation by a maximum-likelihood matching with templates is transferable to other application areas as the offline recognition of symbols. / Obwohl berührungsbasierte Interaktionen mit dem Aufkommen mobiler Geräte zunehmend Verbreitung fanden, beschränken sich Multi-Touch Eingaben größtenteils auf direkte Manipulationen. Im Bereich gestischer Kommandos finden, wenn überhaupt, nur Single-Touch Symbole Anwendung. Der vorliegenden Arbeit liegt der Gedanke zugrunde, dass die Umsetzung von Interaktionstechniken mit der Verfügbarkeit einfach zu handhabender Werkzeuge für deren Interpretation zusammenhängt. Auch kann die Hürde, eigene Techniken zu implementieren, verringert werden, wenn vielfältige Interaktionen erprobt sind und ihr Nutzen für Anwendungsentwickler abschätzbar wird. In der verfassten Dissertation wird ein Erkenner für planare, symbolische Gesten entwickelt, der über die Angabe von Templates trainiert werden kann und keine Beschränkung der Vielfalt von Eingaben auf berührungsempfindlichen Oberflächen voraussetzt. Um eine möglichst flexible Einsetzbarkeit zu gewährleisten, soll die Interpretation einer Geste unabhängig von natürlichen Varianzen - ihrer Translation, Skalierung, Rotation und Geschwindigkeit - und unter wenig spezifizierten Templates pro Klasse möglich sein. Weiterhin sind für Nutzerinteraktionen im Anwendungskontext übliche Echtzeit-Kriterien einzuhalten. Der vorgestellte Gestenerkenner basiert auf der Integration eines Nächste-Nachbar-Verfahrens in einen Ansatz der Bayes\'schen Klassifikation. Gesten werden in elementare, bedeutungstragende Einheiten zerlegt, aus deren lokalen Merkmalen mittels eines Sensor-Fusion Prozesses eine Maximum-Likelihood-Repräsentation abgeleitet wird. Die Flexibilität und hohe Genauigkeit des statistischen Verfahrens wird in ausführlichen Tests nachgewiesen. Unter gleichbleibenden Anforderungen wird eine Erweiterung vorgestellt, die eine Prädiktion von Gesten bei partiellen Eingaben ermöglicht. Deren Nutzen liegt - neben effizienteren Eingaben - in der nachgewiesenen Möglichkeit, per Templates spezifizierte direkte Manipulationen zu interpretieren. Zur Demonstration der Praxistauglichkeit der präsentierten Konzepte werden exemplarisch zwei Anwendungen entwickelt und mit Nutzern getestet, die eine vielseitige Verwendung von Mehr-Finger-Eingaben vorsehen. Neben einem Erkenner trainierbarer, domänenunabhängiger Skizzen wird ein System für die Texteingabe mit den Fingern bereitgestellt. Anhand von Nutzerstudien wird gezeigt, dass Multi-Touch beim Skizzieren verwendet wird, wenn es als Alternative zur Verfügung steht und die Verwendung eines Multi-Touch Gestenalphabetes im Vergleich zur Texteingabe per Single-Touch effizienteres Schreiben zulässt. Von den vorgestellten Konzepten können UI-Designer, Usability-Experten und Entwickler von Feedforward-Mechanismen zum dynamischen Lehren gestischer Eingaben gleichermaßen profitieren. Die Zerlegung einer Eingabe in Token und ihre Interpretation anhand der Zuordnung zu spezifizierten Templates lässt sich weiterhin auf benachbarte Gebiete, etwa die Offline-Erkennung von Symbolen, übertragen.

No description available.

This project was originated by K2 atmitec Brno s.r.o. company. The result is data mining module in K2 information system environment. Engineered data module implements association analysis over the data of K2 information system data warehouse. Analyzed data contains information about sales filed in K2 information system. Module is implementing consumer basket analysis.