Master Thesis MSTR-2020-62

BibliographyGadirov, Hamid: Autoencoder-based feature extraction for ensemble visualization.
University of Stuttgart, Faculty of Computer Science, Electrical Engineering, and Information Technology, Master Thesis No. 62 (2020).
82 pages, english.
Abstract

In this master's thesis, we investigate machine learning methods to support the visualization of ensemble data. Our goal is to develop methods that allow us to efficiently explore the projections of various ensemble datasets and investigate the ability of autoencoder-based techniques to extract high-level data features. This enables clustering of data samples on the projections according to their behavior modes. First, we apply unsupervised feature learning techniques, such as autoencoders or variational autoencoders, to ensemble members for high-level feature extraction. Then, we perform a projection from the extracted feature space for scatterplot visualization. In order to obtain quantitative results, in addition to qualitative, we develop metrics for evaluation of the resulting projections. After that, we use the quantitative results to obtain a set of Pareto efficient models. We evaluate various feature learning methods and projection techniques, and compare their ability of extracting expressive high-level data features. Our results indicate that the learned unsupervised features improve the clustering on the final projections. Autoencoders and (beta-)variational autoencoders with properly selected parameters are capable of extracting high-level features from ensembles. The combination of metrics allow us to evaluate the resulting projections. We summarize our findings by offering practical suggestions for applying autoencoder-based techniques to ensemble data.

Full text and
other links
Volltext
Department(s)University of Stuttgart, Institute of Visualisation and Interactive Systems, Visualisation and Interactive Systems
Superviser(s)Ertl, Prof. Thomas; Frey, Dr. Steffen; Tkachev, Gleb
Entry dateMarch 3, 2021
   Publ. Computer Science