Refine
Document Type
- Master's Thesis (3)
- Bachelor Thesis (1)
- Diploma Thesis (1)
Keywords
- Zeitreihe (5) (remove)
Machine learning models for timeseries have always been a special topic of interest due to their unique data structure. Recently, the introduction of attention improved the capabilities of recurrent neural networks and transformers with respect to their learning tasks such as machine translation. However, these models are usually subsymbolic architectures, making their inner working hard to interpret without comprehensive tools. In contrast, interpretable models such learning vector quantization are more transparent in the ability to interpret their decision process. This thesis tries to merge attention as a machine learning function with learning vector quantization to better handle timeseries data. A design on such a model is proposed and tested with a dataset used in connection with the attention based transformers. Although the proposed model did not yield the expected results, this work outlines improvements for further research on this approach.
Learning Vector Quantization ist ein Klassifikator, der in seiner Urform im euklidischen Raum lernt. Für Zeitreihendaten benötigt es ein gesondertes Distanzmaß, nicht nur wegen der Relation der Zeitpunkte untereinander, sondern auch wegen der unterschiedlichen Längen dieser Zeitreihendaten. Als solches Distanzmaß wird Dynamic Time Warping eingesetzt. Diese Arbeit untersucht die Implementierung und dessen Zeit- und Raumkomplexität.
Prototype-based classification methods like Generalized Matrix Learning Vector Quantization (GMLVQ) are simple and easy to implement. An appropriate choice of the activation function plays an important role in the performance of (deep) multilayer perceptrons (MLP) that rely on a non-linearity for classification and regression learning. In this thesis, successful candidates of non-linear activation functions are investigated which are known for MLPs for application in GMLVQ to realize a non-linear mapping. The influence of the non-linear activation functions on the performance of the model with respect to accuracy, convergence rate are analyzed and experimental results are documented.
In the following study we evaluated capabilities of how a simple autoencoder can be used to trainGeneralized Learning Vector Quantization classifier. Specifically, we proved that the bottlenecks of an autoencoder serve as an "information filter" which tries to best represent the desired output in that particular layer in the statistical sense of mutual information.
Autoencoder model was trained for purely unsupervised task and leveraged the advantages by learning feature representations. As a result, the model got the significant value of the accuracy. Implementation and tuning of the model was carried out using Tensor Flow [1].
An extra study has been dedicated to improve traditional GLVQ algorithm taken from sklearn-lvg [2] using the bottleneck from an autoencoder.
The study has revealed potential of bottlenecks of an autoencoder as pre-processing tool in improving the accuracy of GLVQ. Specifically, the model was capable to identify 75% improvements of accuracy in GLVQ comparing to original one, which has about 62%. Consequently, the research exposed the need for further improvement of the model in the present problem case.
Ziel der Diplomarbeit ist es, ein System zur Darstellung von mehrdimensionalen Zeitreihen zu entwickeln, welches zur Visualisierung der Ergebnisse von Sensoren zur Qualitätskontrolle an Laserschweißnähten eingesetzt werden soll. In dieser Arbeit wird zuerst ein Überblick über Qualitätskontrollsysteme in der Laserprozesstechnik vermittelt. Neben aktuell verwendeten Sensoren zur Gewinnung der Daten werden die Grundlagen der Visualisierung präsentiert und die Grenzen des aktuellen Visualisierungstools aufgezeigt. Nach der Auswahl einer Toolbox werden systematische Untersuchungen in Bezug auf Performance und Funktion durchgeführt. In weiteren Kapiteln werden Design, Implementierung und Verifikation des neu entwickelten und auf der gewählten Toolbox basierenden Visualisierungssystems präsentiert. Die Zusammenfassung der Ergebnisse sowie ein Ausblick auf mögliche Optimierungen schließen diese Arbeit ab.