Refine
Year of publication
Document Type
- conference proceeding (article) (21)
- Part of a Book (8)
- Article (2)
Reviewed
- Begutachtet/Reviewed (17)
Keywords
- Automotive Security (2)
- software engineering (2)
- Automotive (1)
- Automotive Software Engineering (1)
- Automotive System Architecture (1)
- Automotive engineering , Software engineering , Industries , Education , Conferences , Machine learning , Automobiles (1)
- Autonomous Driving (1)
- Cloud Attacks (1)
- Dependability (1)
- Machine Learning (1)
Institute
SafeML: Safety Monitoring of Machine Learning Classifiers Through Statistical Difference Measures
(2020)
Variability management has become an important concern in software and systems engineering. Especially in industrial settings a rigid management of variability has been identified as an important prerequisite for further optimization of the development process, e.g. for reuse of software sub-systems across vehicle models such as the Mercedes Benz A-Class and C-Class. In response to this growing practical interest, the scientific community has come up with numerous concepts and techniques for modeling variability. However, despite initial attempts to integrate or unify some of these manifold approaches, a clear understanding of how they precisely relate to each other is still not yet achieved.
In the paper, various techniques for variability modeling are elaborated and a basic classification scheme is proposed. From this we derive their common capabilities, which arguably embody the essence of variability modeling in general. On this basis, a discussion is presented that concerns the potential and feasibility of integrating all these diverse techniques into a single, common technique for variability modeling.
The increasing use of software and connectivity in modern vehicles has made cybersecurity an important issue in the automotive industry. ISO 21434 is a standard for automotive cybersecurity engineering that provides guidelines for the development and validation of secure automotive systems. For effective implementation and practical use of ISO 21434, it must be incorporated into existing automotive industry development workflows.
In this paper, we investigate the practical applicability of ISO 21434 in the context of the Security Abstraction Model (SAM), a security modeling approach for the domain-specific modeling language EAST-ADL, and provide insights into the benefits and consequences of this approach. In doing so, we describe the methodological opportunities of integrating ISO 21434 into SAM on the one hand and present a case study illustrating the application of this integrated approach in the development of a secure automotive system on the other hand. Our results suggest that the integration of ISO 21434 into SAM better supports automotive system security in the early development phases and makes it transparent to a wide range of stakeholders. At the same time, it becomes clear that a representation of the interrelationships in the form of a metamodel, in contrast to ISO 21434 in which these are only described textually, significantly improves conceptual understanding and ultimately enables pragmatic usability in industrial development.
Evaluation and modeling of the supercore parallelization pattern in automotive real-time systems
(2019)
Machine learning classifiers for image recognition are prevalent in many applications. We study the problem of finding adversarial examples for such classifiers, i.e., to manipulate the images in such a way that they still look like the original images to a human but are misinterpreted by the classifier. Finding adversarial examples corresponds to a search problem in the image space. We focus on black-box attacks that can only use the original classifier to guide the search. The challenge is not to find adversarial examples, but rather to find them efficiently, ideally in real time. We show two novel methods that increase the efficiency of black-box search algorithms for adversarial examples: The first uses a relevance mask, i.e., a bitmask on the original image that restricts the search to those pixels that appear to be more relevant to the attacked classifier than others. The second exploits the discovery of merge drift, a phenomenon that negatively affects search algorithms that are based on the merging of image candidates. We evaluate both concepts on existing and new algorithms.