Compositional Falsification of Cyber-Physical Systems with Machine Learning Components
This addresses safety concerns for automotive systems by providing a method to detect failures, though it is incremental as it builds on existing falsification techniques.
The paper tackles the problem of verifying cyber-physical systems with machine learning components by formulating it as a falsification problem using signal temporal logic, and demonstrates efficacy on an automatic emergency braking system with deep neural networks.
Cyber-physical systems (CPS), such as automotive systems, are starting to include sophisticated machine learning (ML) components. Their correctness, therefore, depends on properties of the inner ML modules. While learning algorithms aim to generalize from examples, they are only as good as the examples provided, and recent efforts have shown that they can produce inconsistent output under small adversarial perturbations. This raises the question: can the output from learning components can lead to a failure of the entire CPS? In this work, we address this question by formulating it as a problem of falsifying signal temporal logic (STL) specifications for CPS with ML components. We propose a compositional falsification framework where a temporal logic falsifier and a machine learning analyzer cooperate with the aim of finding falsifying executions of the considered model. The efficacy of the proposed technique is shown on an automatic emergency braking system model with a perception component based on deep neural networks.