The problem
Search-based testing is widely used to find bugs in models of complex cyber-physical systems. Recent work casts it as the falsification of formally specified temporal properties, using the robustness semantics of Signal Temporal Logic.
Scaling this to highly complex engineering systems requires efficient falsification procedures that also work on black box models, and the task is made harder by inputs that are often time-dependent functions.
The approach
Falsification of black box cyber-physical systems with techniques from active learning, tailored to time-dependent, functional inputs.
Results
A considerable gain in computational effort, obtained by reducing the number of model simulations needed.