D2.1 Existing AI Algorithms and their Accountability and Resilience Features within the Context of Applications to IoT, 5G, and Cybersecurity
The present document seeks to understand the accountability and resilience of existing Machine Learning (ML) algorithms to establish a foundation for the research activities towards a SPATIAL platform. The selection of ML algorithms analysed in this document is based on their potential application in the four SPATIAL use cases, which reflect the domains Internet of Things (IoT), 5G, cybersecurity, and eHealth. In order to acquire the above-mentioned understanding, this deliverable analyses the identified relevant ML algorithms concerning their accountability, explainability and resilience characteristics. Different Explainable AI (XAI) methods are considered, and their applicability to the ML algorithms in question is discussed. Furthermore, the selected ML algorithms are looked into with regard to their resilience against modern attacks on the AI models (e.g., data poisoning, model stealing, evasion attacks etc.).
Read the deliverable here!