Privacy Protection-considered Task Offloading In Mobile Edge Computing

Mobile edge computing is a new network architecture that migrates application services and computing resources to the side of user and provides task offloading services to mobile terminals.Different from traditional cloud computing,application tasks of mobile terminals can be offloaded to nearby edge servers,reducing network congestion and computing costs caused by long-distance task transmission,and effectively improving the quality of user experience.However,in the process of task offloading,the task offloading preference of the mobile terminal to reduce the computing delay and energy consumption can expose the user’s privacy,and the adversary can monitor the offloading task to infer the user’s location and device usage mode.If the task offloading policy does not protect user privacy,it will seriously affect the enthusiasm of users to participate in mobile edge computing.The definition of privacy protection level in related research on task offload preference leakage privacy has certain flaws,and when the privacy protection level converges to a fixed value,user privacy can still be inferred from the fixed value and task offloading amount.In addition,related work focus on mobile edge computing networks with single access points,and the privacy risks of multi-access point joint speculation of user location have yet to be explored.In this dissertation,we conduct research on privacy protection-considered task offloading in mobile edge computing,so as to solve the problem of user privacy leakage caused by offloading preference.We focus on two types of privacy issues: user location and device usage patterns,and the basic idea is: from the perspective of probability,the information entropy is used to describe the randomness of the information leaked by the current offloading decision,and the information more random,the more difficult it is to infer user information.On one hand,we propose an offloading scheme considering privacy protection with single access point based on deep reinforcement learning.In the mobile edge computing network with a single access point,by comparing the similarity of current offloading decision and history offloading decision,the privacy entropy is defined to describe the chaos of the channel state information contained in the current offloading decision,and the privacy level constraint is added to the existing task offloading model to establish a sequential decision problem,which is expressed as the Markov decision process.Considering the time-varying channel gain and random task arrival rate,an SAC-PP algorithm based on deep reinforcement learning is proposed to minimize the weighted sum of privacy level and computational cost of mobile devices.On the other hand,we propose an offloading scheme considering privacy protection with multiple access points based on stochastic game.In a mobile edge computing network with multiple access points,the privacy problem of multiple access points jointly guessing the user’s location is considered,and the privacy protection level is evaluated by defining the privacy entropy of mobile devices for offloading preferences to different edge servers.We fully consider the mobility of terminal devices during task offloading,and jointly optimizes offloading decisions and resource allocation to ensure user privacy and quality of experience.In addition,considering the problems of uplink channel interference and edge node computing resource contention caused by multiple users,the task offloading optimization problem is described as a stochastic game,and by establishing a trusted third party,the policy network is trained centrally under the premise of protecting user privacy to reduce the environmental instability,and an algorithm based on multi-agent reinforcement learning is proposed to solve the optimal strategy.