Confidentiality and privacy on constrained devices: optimization using heuristic and machine learning algorithms

Abstract

Due to the high frequency of data flow for IoT devices, it becomes a necessity to take advantage of cloud capabilities to store these data and respond to high volumes of users’ demand. Statistics show that the number of IoT devices will reach fifty billions by 2020 Taha et al. (2019a). IoT applications are backed through clouds where data are stored and processed by gigantic processing systems and are accessed only by authorized users. Cloud is honest but still curious. In this context, data confidentiality on Cloud servers must be protected, so as to ensure that only authorized users can access it. This also entails that even Cloud service providers, who host the data, should not reveal it. To this end, input data must be first encrypted on IoT devices before being uploaded to the Cloud which will, in turn, add computation overhead on IoT devices. In this thesis, a new scheme is proposed to reduce the overhead of Ciphertext Policy Attribute Based Encryption (CP-ABE) intensive tasks on constrained devices. This scheme solves all the aforementioned challenges by considering the following phases: (1) The scheme can smartly be adapted from full CP-ABE encryption to partial CP-ABE encryption based on the context of the constrained device, the complexity of the task, and the access policy. It uses machine learning technique to decide when it should perform full or partial CP-ABE encryption. (2) The scheme considers the mobility feature of the constrained devices (i.e., OBU in VANET network). Hence, it can minimize the intensive computation of CP-ABE tasks in order to reduce the execution time by dividing and distributing it on the cluster vehicles. We use Kuberents to build the cluster of vehicles and prepare the infrastructure needed to run CP-ABE tasks. (3) The scheme considers the multi-objective optimization and minimizes the time of the encryption task. We take into account the limitations of the available resources in dynamic topology such as VANET. Hence, the scheme minimizes the number of resources used to perform CP-ABE tasks and reduces the time of the CP-ABE encryption operations itself. Finally, in order to achieve our objectives, we use machine learning technique to divide the task into sub-tasks, taking into consideration the factors that impact the computation overhead. We also use Particle Swarm Optimization (PSO) in our approach to perform tasks distribution process.

Date	10 Feb 2020
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Chamseddine Talhi (Supervisor)