[BA] Variability-aware Runtime Visualization for Expensive Blackbox Optimization

Expensive Blackbox Optimization~(EBBO) problems are widely-spread in industry and research in domains such as: machine learning, search-based software engineering, operation research and many others. Runtime visualization of an EBBO experiment provides insights on the optimization process and can potentially serve as a basis for triggering runtime reconfiguration.

BRISE-MPL is a multiple product line for expensive blackbox optimization, developed at the chair of software technology. It offers a wide spectrum of configuration options, enabling to tailor the optimization algorithm to the problem at hand. Unlike the optimization, the visualization capabilities of the framework are underachieving. The main deficiency is missing variability-awareness of the Web front-end component, which is responsible for runtime visualization. It results in a limited number of visualization options and their customizability. Moreover, it is not possible to neither pre-select a visualization option for a specific optimization problem instance nor to restrict its usage.

The goal of this thesis is to enable variability of runtime visualization in BRISE-MPL. To reach this goal the following tasks need to be accomplished.

• Analyze BRISE-MPL with a special focus on runtime visualization and perform requirements engineering.
• Examine state-of-the-art approaches for variability management and visualization techniques for EBBO.
• Propose an extension of the \texttt{Web front-end} component, enhancing its variability-awareness.
• Implement the proposed approach and integrate it into BRISE-MPL.
• Evaluate the novel approach against the existing Web front-end component.

Betreuer: Dmytro Pukhkaiev