MCollections——17

MCollections——17

Differential Privacy Oriented Distributed Online Learning for Mobile Social Video Prefetching

面向差异隐私的分布式在线学习的移动社交视频预取

Today the ever fast growing mobile social video traffic motivates the urgent requirement for new designs alleviating the backbone pressures while ensuring the user-quality experience.Mobile video prefetching gearing towards traffic offloading and delay reduction can be a promising solution. Howbeit, providing high performance prefetching remains problematic in the presence of the high dynamic mobile users’ viewing behaviors and consecutive generated video content. Furthermore, making prefetching decision requires sensitive viewing history involving increasing privacy issues and the frequent occurrence of personal information leakage should be concerned about as well. Among the extensive studies on privacy preserving, differential privacy receives much attention that blocks the accurate inference from the output and perturb the published data without influencing the overall data features. Differential privacy mechanism has been considered into several perspectives of the video sharing system, nonetheless, no existing prefetching solution has taken privacy preserving into consideration while ensures the system performance.

In this paper, the authors propose a differential privacy oriented distributed online learning method for mobile social video prefetching (DPDL-SVP). They discover that users’ viewing behaviors have strong connection with video preference, content popularity, and social interactions. Therefore, they formulate an online convex optimization problem taking the above three factors into consideration, divide the above problem into two subproblems, and implement a distributed algorithm sequentially to solve them with differential privacy. Performance bound of the proposed online algorithms has been theoretically provided and verified via a series simulation, besides, the authors implement the algorithm through real viewing history traces to evaluate its performance.

 Fig. Overview of distributed learning-based mobile social video prefetching