群眾感知系統和伺服器為主激勵機制

Abstract

群眾感知是一種利用無處不在的智能手機來收集人類活動和周邊環境的方法。在群眾感知系統裡，感測任務中的大量用戶藉由他們的行動裝置來收集資料並將資料傳送給資料收集伺服器。系統的性能大部分取決於大眾的參與，因此，激勵機制在群眾感知系統裡是非常重要的。在此篇論文裡，我們著重於伺服器為主的激勵機制設計。我們設計四種使用斯塔克爾伯格模型的激勵機制，其中伺服器是領主導者，使用者們是跟隨者。我們研究兩種獎勵方式，第一種是先前論文提到的:總和固定的獎勵，第二種是總獎勵與所花費的時間或收集的信息成比例。而獎勵分配的方式是依據使用者所花的時間或是所收集到的資訊量來按比例分配。我們先假設使用者收集資料的能力是一樣的，但資源的價值對每個使用者來說或許是不同的。基於以上的獎勵和獎勵分配方式我們有TR-T, TR-Q, DR-T, DR-Q 四種模型。我們研究同異質使用者的情況。在同質使用者的情況中，我們在TR-T, TR-Q, DR-Q模型裡證明了純策略斯塔克爾伯格均衡的存在和唯一性。在異質使用者的情況中，TR-T 和DR-Q 模型有唯一的斯塔克爾伯格均衡。最後我們使用PoA和PoS來分析整個系統的效能。除了少數在TR-T, TR-Q的情況，PoA的結果可以被常數界定。

Crowdsensing is an approach to collect human activities and surrounding environment which takes the benefit of the pervasive smartphones and their powerful sensors. In a crowdsensing system, a large number of users in the sensing tasks collect and send data through their mobile devices to a data collection server. The performance of the system heavily depends on the crowd participation. Thus, incentive mechanisms are important in crowdsensing. We focus on the server-centric model, in which the server has more control over the payment. We design four incentive mechanisms using Stackelberg game, where the server is the leader while the users are the followers. We study the original scenario in which the total reward is fixed and a different scenario in which the total reward is proportional to the effort spent or information collected. The reward is distributed to the users in proportion to either the amount of time spent or quantity of information. We assume that the ability of information collect is the same for all users, but the value of resources may be different. Based on the above reward models and distribution methods, we formulate four different models called TR-T, TR-Q, DR-T, DR-Q models respectively. We study the cases with homogeneous and heterogeneous users. For homogeneous users, we can prove the existence and uniqueness of pure strategy Stackelberg equilibrium in TR-T, TR-Q, DR-Q models. For heterogeneous users, there is a unique Stackelberg equilibrium in TR-T and DR-Q models. We compute the efficiency which measured by PoA and PoS for homogeneous models and DR-Q model with heterogeneous users. The PoA is bounded by a constant except for some special cases in TR-T and TR-Q models.