Increasing the Confidence of Deep Neural Networks by Coverage Analysis
This addresses safety-critical applications like autonomous vehicles by improving model robustness, though it appears incremental as it builds on existing coverage paradigms.
The paper tackles the problem of making deep neural networks more trustworthy and robust against various unsafe inputs, such as adversarial examples and out-of-distribution data, by proposing a lightweight monitoring architecture based on coverage analysis. The result shows that the approach effectively detects these inputs with limited extra execution time and memory requirements.
The great performance of machine learning algorithms and deep neural networks in several perception and control tasks is pushing the industry to adopt such technologies in safety-critical applications, as autonomous robots and self-driving vehicles. At present, however, several issues need to be solved to make deep learning methods more trustworthy, predictable, safe, and secure against adversarial attacks. Although several methods have been proposed to improve the trustworthiness of deep neural networks, most of them are tailored for specific classes of adversarial examples, hence failing to detect other corner cases or unsafe inputs that heavily deviate from the training samples. This paper presents a lightweight monitoring architecture based on coverage paradigms to enhance the model robustness against different unsafe inputs. In particular, four coverage analysis methods are proposed and tested in the architecture for evaluating multiple detection logics. Experimental results show that the proposed approach is effective in detecting both powerful adversarial examples and out-of-distribution inputs, introducing limited extra-execution time and memory requirements.