In recent years, the Internet of Things (IoT) has progressed dramatically due to advancements in technology. It plays an important role in the development of abundant applications and economic ventures. In IoT, smart devices communicate with each other using the Internet without any human intervention. The 6G vision also considers the massive IoT as a driving force, in which a strong relationship has been identified between 6G and high-performance mobile edge computing . While edge computing resources will handle some of the IoT and mobile device data, much of it will require more centralized resources to perform the data processing task.
To execute the big data analytics generated by IoT devices easy, the whole IoT network can be divided into different subnetworks and data of each subnetwork is collected, aggregated and forwarded by their respective central nodes. In such cases, the presence of malicious nodes causes severe issues in the sensing results, localization and service provisioning, which discourages new entities to join the network. Therefore, it is very important to establish trust between all entities by detecting and removing such nodes. Moreover, for reliable service provisioning in IoTs, the service provider nodes deliver services to the client nodes. However, there is no mechanism to collect enough information that assures the non-repudiation of both service provider and client node in the service provisioning mechanism. A blockchain-based localization mechanism is proposed for resource-constrained IoT nodes to find their location . However, there is no mechanism yet to prevent the client nodes from repudiating about actually demanded services.
The federated learning for the detection of malicious nodes will be further studied, which uses SVM and RF classifiers for identifying malicious nodes . In terms of implementation, the data of the IoTs network is initially collected by the respective sink node and then provided to the virtual machine associated with it. The model on this virtual machine is trained using the distributed dataset. After the models’ training, the trained models are sent to the B5G fog server. The B5G fog server fuses all the models and sends the fused model back to each virtual machine. Then these virtual machines use this model for the classification of legitimate and malicious nodes of their clusters.
The project aims to investigate the distributed AI and implement the federated learning techniques in the dense IoT network in which the malicious nodes are detected in the network without compromising the privacy of sensing data. Federated Learning enables the IoT's data to remain on the device, the model will be further investigated to articulate how it learns with time and with the collaborative effort of all distributed IoT devices as agents.
Successful completion of the project requires addressing the following technical objectives.
- Analyse and simulate the distribution of IoT networks, sensing data and security log files.
- Design and compare various distributed AI algorithms to detect malicious users in the IoT network and measure their efficiency.
- Investigate the reduction in latency due to the generated learning model remaining on the device and the dependency on the central cluster is reduced.
- Investigate the proposed algorithms in terms of energy efficiency since IoT devices have very limited processing capacity, memory and power source.
Applicants should have, at least, an Honours Degree at 2.1 or above (or equivalent) in Computer Science or related disciplines. In addition, they should have a good mathematical background, excellent programming skills in Python/MATLAB and an interest in machine learning.