Happy to announce our newest paper published in the proceedings of UR-AI 2021 conference
Abstract: Embedded systems are the basis for many electronic devices. As a combination of hardware and software designed for specific purposes, Embedded Systems ensure the functionality of Connected Cars, Autonomous Driving, Smart Farming, Industrial Internet of Things and Smart Homes. The enormous competitive pressure forces manufacturers to significantly shorten their time to market and thus reduces the corresponding production cycles. This challenge is directed to the same extent to quality assurance. Due to the constantly growing number of (regression) tests, it is no longer practicable to perform all verifications in all development phases up to the finished product: each quality feature is planned and configured individually. But this approach is usually carried out manually with a lot of effort and is rather rarely adapted over time. On the other hand, software changes very quickly: new features are added, new dependencies arise or are resolved. Communication between individual components change. The probability that errors are found by tests (too) late is substantially increased with each change.
This paper presents an approach that successfully mitigates this challenge with the help of suitable Artificial Intelligence methods. To reduce lead time, a mechanism is developed that reduces the number of required tests. Based on the data from previous verifications, a (significantly) smaller subset of tests, which is sufficient to verify the correctness of the change, is selected. The remaining probability that tests necessary for negative verification of the software are not considered, is thereby accepted. Initial results, using data from several open-source projects as well as the use of a prototype machine learning pipeline, show promising results with respect to their predictive capabilities.