Master Theses in Visual Data Science @ VRVis Research Center, 2018

Type: 
DA
Persons: 
5
Workgroup: 

Description

The Visual Analytics group at the research center VRVis in Vienna (close to the U1 station "Kaisermühlen") is looking for your support! We tackle visual data science challenges in a friendly and inspiring atmosphere. We are offering several student positions (f/m) to conduct your master thesis research. You are the right fit, if you would like your master thesis to be:

  • at the heart of visual data science
  • useful/ applicable, i.e. directly addressing practical needs of notable industry partners in Industry 4.0 (e.g. RHI Magnesita), the automotive industry (e.g. AVL List), and the energy sector (e.g. Austrian Power Grid)
  • using real data for the benefit and in collaboration with domain experts from our industry partners
  • contributing to our large-scale high-performance software system "Visplore" (http://goo.gl/wqJ4AS) which is applied by more than 10 companies
  • a great opportunity to extend skills and gain practical experience in data science, software engineering, C++, multi-threaded programming, and Visual Analytics
  • tightly integrated into our effort to offer innovative solutions to practical challenges of visual data science
  • well supported by our experienced and motivated team
  • financially compensated upon successful completion

For more information about our research group please refer to: https://www.vrvis.at/research/visual-analytics-group/
A video illustrating our software Visplore for industrial applications can be found at: https://youtu.be/fpyDfj9sUjk

Topics

Pattern search in industrial process data

PatternIn Industry 4.0, a huge number of sensors permanently acquires data about the states of machinery and environmental conditions. Using this data for tasks such as predictive maintenance is a key goal in today’s industry.

This topic is about combining state-of-the-art methods from time series mining with methods from Visual Analytics such as query-by-example. Users should be able to describe and automatically detect anomalies and critical process states in a large number of long industrial time series. In addition to the algorithms, designing suitable user interfaces for specifying search patterns and presenting results are important aspects of the topic.

Smart view recommendation for fast insights
When exploring new data, it is often not known in advance, which visualizations provide the most interesting insights. The aim of this topic is to shorten the time for tedious manual identification of interesting views on industrial process data by suggesting a small number of interesting views to the user. These views may reveal outliers, for example, as well as unexpected distributions, relationships, and trends in the data. The topic requires integrating data mining and machine learning techniques for an automated search of interesting patterns with an intuitive overview of the possible options.

Streamlined visual workflows in data science platforms
Data scientists commonly work in environments such as Python or R, using tools such as Spyder, Jupyter notebooks, or R Studio. An acceptance of new interactive visualization techniques by data scientists highly depends on the ease at which these techniques can be accessed from within their environments. Besides bi-directional data exchange, the integration also involves enabling access to advanced machine learning techniques (e.g. from Scikit learn) directly from interactive dashboards for building and validating models. This topic largely builds on existing implementations of the integration (e.g. Python is used within Visplore) and has the goal of taking possible use cases one step furhter.

Exploration and interactive feature extraction from cyclic data
Many time series in industry, medicine, and other sectors consist of a sequence of cyclic patterns, e.g., originating from temporal cycles or batch production. The aim of this diploma thesis is to create an interactive dashboard based on the software Visplore that allows to extract cycles from long time series. Interactive views should then support to compare cycles and provide interactive tools for extracting relevant features (e.g., how long a cycle exceeds a user-defined threshold).

 

Requirements

Candidates should have experience with object-oriented programming (ideally C++), some familiarity with machine learning methods, knowledge of software engineering methods, and should be interested in data science.

Contact @ VRVis

Please send your CV and a short motivational letter including your preferred diploma thesis
topic(s) to Mr. Harald Piringer. hp@vrvis.at

Applications are always welcome.

We would like to especially encourage female candidates to apply!

Please don't hesitate to get in touch with us.

Contact

For more information please contact Harald Piringer (hp@vrvis.at) or Eduard Gröller (groeller@cg.tuwien.ac.at).