Faeq Alrimawi

Amel Bennaceur, Gopi Krishnan Rajbahadur, Prince Mercy et al.

Organisations designing, developing, and deploying machine learning systems (MLS) need to be able to check that these systems are trustworthy, and communicate this clearly to their stakeholders, be they different categories of users, engineers, or wider society. By focusing on stakeholders, Requirements Engineering is well positioned to drive the design and engineering of MLS that align with the needs of their stakeholders. Yet, we still need a systematic process for modelling and reasoning about requirements for MLS that is driven both by stakeholders' needs and constraints for MLS development. This paper proposes a framework entitled REAL (Requirements Engineering for mAchines that Learn - and Fail) to help develop MLS that align with stakeholders' needs by adopting a requirements engineering approach. This model-based framework is based on three principles. First, weaving together requirements for data, models, and the system as a whole. Second, using failure to drive the exploration of alternative requirements. Third, iterative and traceable refinement of MLS requirements. We demonstrate the proposed framework using an example from autonomous driving and show that REAL supports the development of MLS that better align with stakeholders' requirements. A replication package is available online.

Faeq Alrimawi, Liliana Pasquale, Deepak Mehta et al.

Cyber-physical systems (CPSs) are part of most critical infrastructures such as industrial automation and transportation systems. Thus, security incidents targeting CPSs can have disruptive consequences to assets and people. As prior incidents tend to re-occur, sharing knowledge about these incidents can help organizations be more prepared to prevent, mitigate or investigate future incidents. This paper proposes a novel approach to enable representation and sharing of knowledge about CPS incidents across different organizations. To support sharing, we represent incident knowledge (incident patterns) capturing incident characteristics that can manifest again, such as incident activities or vulnerabilities exploited by offenders. Incident patterns are a more abstract representation of specific incident instances and, thus, are general enough to be applicable to various systems - different than the one in which the incident occurred. They can also avoid disclosing potentially sensitive information about an organization's assets and resources. We provide an automated technique to extract an incident pattern from a specific incident instance. To understand how an incident pattern can manifest again in other cyber-physical systems, we also provide an automated technique to instantiate incident patterns to specific systems. We demonstrate the feasibility of our approach in the application domain of smart buildings. We evaluate correctness, scalability, and performance using two substantive scenarios inspired by real-world systems and incidents.