Secured Frank Wolfe learning and Dirichlet Gaussian Vicinity based authentication for IoT edge computing

Abstract

With the evolution of the Internet of Things (IoT) several users take part in different applications via sensors. The foremost confront here remains in selecting the most confidential users or sensors in the edge computing system of the IoT. Here, both the end-users and the edge servers are likely to be malicious or compromised sensors. Several works have been contributed to identifying and isolating the malicious end-users or edge servers. Our work concentrates on the security aspects of edge servers of IoT. The Frank-Wolfe Optimal Service Requests (FWOSR) algorithm is utilized to evaluate the boundaries or limits of the logistic regression model, in which the convex problem under a linear approximation is solved for weight sparsity (i.e. several user requests competing for closest edge server) to avoid over-fitting in the supervised machine learning process. We design a Frank Wolfe Supervised Machine Learning (FWSL) technique to choose an optimal edge server and further minimize the computational and communication costs between the user requests and the edge server. Next, Dirichlet Gaussian Blocked Gibbs Vicinity-based Authentication model for location-based services in Cloud networks is proposed. Here, the vicinity-based authentication is implemented based on Received Signal Strength Indicators (RSSI), MAC address and packet arrival time. With this, the authentication accuracy is improved by introducing the Gaussian function in the vicinity test and provides flexible vicinity range control by taking into account multiple locations. Simulation and experiment are also conducted to validate the computational cost, communication cost, time complexity and detection error rate.