Applying Deep Learning For Imitating Adaptive Agent Behavior in Statistical Software Testing

André Reichstaller, Benedikt Eberhardinger, Hella Seebach, Alexander Knapp, and Wolfgang Reif

Statistical test generation builds on profiles which describe the estimated conditions of the system under test’s environment. Such environmental profiles, however, do not directly provide us with inputs for testing particular system components, as those mostly depend on the output of others. We thus additionally need to estimate this output if we want to maintain statistical accuracy. Instantiating this task for the isolated testing of self-organization mechanisms between adaptive agents, this paper investigates the application of deep learning techniques for imitating the agents’ output. The proposed technique is evaluated on a simulated self-organizing grid of power plants.
Softwaretechnik-Trends 37:3