Optimal and Privacy-Aware Resource Management in Artificial Intelligence of Things Using Osmotic Computing

Critical infrastructure comprising on-demand devices, including secondary servers, comes into play when a situation like an overload is involved. The on-demand servers and devices require smart management solutions that form an integral part of Artificial Intelligence of Things (AIoT). This work considers AIoT as a combination of Mobile-Internet of Things (M-IoT) and AI requiring immediate response, secondary support system, and computational resources. Privacy in AIoT is always a concern when sharing information as intruders can eavesdrop on the settings of the system. This article uses an osmotic computing paradigm, which enables the derivation of strategies to decide on the methods of sharing services via optimal and privacy-aware resource management in AIoT. A safety competition is built on top of configuration rewards that help to attain privacy-by-design. The contributions of this article are expressed using theoretical analysis and numerical simulations.

I. Introduction

Resource management ensures that the network is available to handle unscheduled or emergency tasks required for supporting excessive service requests, setup of rescue operations, and even deploying on-demand nodes facilitated by edge computing [1], [2]. With the pressure of connecting billions of intelligent devices and having mobile terminals as a part of the Internet of Things (IoT), the intensity of migrating the services and assuring the safety of operations is critical [3]. Most recent works focus on service migration when an additional load is encountered in the network or in the case of a change in the configuration of the network [4], [5]. These works utilize cost functions and build a priority system to handle incoming requests. This facilitation is covered as a part of resource management. The management of resources demands high-quality operations, better control, security, and privacy which withholds all the information that may help an intruder to pick its target based on the service requests [6], [7].