Abstract

Data sharing and federation can significantly increase efficiency and lower the cost of digital collaborations. It is important to convince the data owners that their outsourced data will be used in a secure and controlled manner. To achieve this goal, constructing a policy governing concrete data usage rule among all parties is essential. More importantly, we need to establish digital infrastructures that can enforce the policy.
In this thesis, we investigate how to select optimal application-tailored infrastructures and enhance policy compliance capabilities. First, we introduce a component linking the policy to the infrastructure patterns. The mechanism selects digital infrastructure patterns that satisfy the collaboration request to a maximal degree by modelling and closeness identification. Second, we present a threat-analysis driven risk assessment framework. The framework quantitatively assesses the remaining risk of an application delegated to digital infrastructure. The optimal digital infrastructure for a specific data federation application is the one which can support the requested collaboration model and provides the best security guarantee. Finally, we present a distributed architecture that detects policy compliance when an algorithm executes on the data. A profile and an IDS model are built for each containerized algorithm and are distributed to endpoint execution platforms via a secure channel. Syscall traces are monitored and analysed in endpoint points platforms. The machine learning based IDS is retrained periodically to increase generalization. A sanitization algorithm is implemented to filter out malicious samples to further defend the architecture against adversarial machine learning attacks.