Overview
In a first step, a deeper understanding and the motivation for "Anomaly Detection" should be created. Proven techniques will be briefly examined and the advantages and potential of Gaussian processes will be shown.
In a second step, with the help of appropriate literature, a clear understanding of the theoretical foundations necessary for the application of Gaussian processes is created. The theory is explained step by step with simple examples and the associated code snippets.
Building on these fundamentals, it will be shown how Gaussian processes can be linked and built up into deep Gaussian processes. For this purpose, the properties resulting from this concatenation are to be shown.
In the next step, methods of dimensionality reduction with the help of Gaussian processes will be shown. For this purpose, the methods GP-LVM, DGP Autoencoders and DGP Variational Autoencoders will be introduced and explained.
Finally, these dimensionality reduction methods will be applied to a Roche sensor data set and compared to conventional techniques such as PCA, MDS and t-SNE
