|16:05||Aleksandar Dimovski||A Decision Tree Lifted Domain for Analyzing Program Families with Numerical Features|
|16:35||Philipp Chrszon||Role-based Automata: Modeling and Formal Analysis of Context-Dependent Systems|
|17:10||Davide Basile||Static Detection of Equivalent Mutants in Real-Time Model-based Mutation Testing|
|17:40||Virtual coffee break|
|18:00||Norbert Siegmund||Keynote Modelling the Universe: Accurate & Interpretable Performance Models for an Astronomical Number of Influences|
Norbert SiegmundUniversity of Leipzig, Germany
"Modelling the Universe: Accurate & Interpretable Performance Models for an Astronomical Number of Influences"
Abstract. Nowadays, nearly all software systems provide configuration capabilities to the user that enable to tune quantitative aspects of the system, such as performance and energy consumption. However, due to the exponential number of configurations rising from the available configuration options, developers, administrators, and users alike are overwhelmed by an astronomical number of possible influences affecting the system's properties. To support the selection of suitable and optimal configurations, several sampling and learning approaches have been proposed in recent years to tame the complexity of the configuration space. In this talk, I will discuss characteristics of performance and how it affects learning of a performance influence model. I will show that different learning techniques have distinct benefits and drawbacks and especially discuss the tension between accuracy, interpretability, and correctness. Finally, I give some ideas on how to address the scalability problem of the exponential configuration space for learning.