بهینه سازی سود در بازار داده های سرویس گرا
Abstract
1- Introduction
2- Related work
3- Market model and problem description
4- Optimal pricing mechanism
5- Numerical results
6- Conclusion
References
Abstract
Deep learning has drawn a lot of attention recently. It is successful in a variety of applications from natural language processing to autonomous vehicle control. A main difference from traditional learning is that deep learning learns the representations of the data, i.e., the features, via greedy layer-wise pre-training. The characteristics of deep learning promotes it as a power tool to mine the data from Internet of Vehicles (IoV). The paper focuses on the economic aspect of IoV and investigates deep learning enabled IoV market for data trading and processing. The economic model of IoV consists of three side: the data provider, the service provider, and the user. The data provider collects the data for the user. The user buys the raw data. The data is further processed by the service provider, who provides the learned features for the user to obtain some profit. To optimize the profit of three-sided participators, a Stackelberg game is proposed to model the interactions among them. We derive the equilibrium pricing mechanism of the providers and corresponding demands of the users. The existence and the uniqueness of the equilibrium strategies are proved. Our analysis reveals that the strategy of each participator is related to the utility of the user and the data/service provider’s cost. To the best of our knowledge, this is the first time that the data provider and the service provider directly interact with the user in the data market.
Introduction
With the rapid development of wireless communication, mobile computing, sensor techniques, and mobile social networking, many objects in our daily lives are interconnected through the Internet. The past few years have witnessed the proliferation of Internet of Things (IoT) paradigm [1–5]. Via smartphones, diverse sensors/actuators, Radio Frequency IDentification (RFID) tags, an enormous amount of data from different aspects and sectors is generated everyday. For instance, it is reported Google approximately processes 20,000 TB data and Flicker approximately generates about 3.6 TB data everyday [6]. It is expected that by 2020, the total amount of data worldwide will reach 35 ZB [7]. The services offered by IoT paradigm highly depend on the available data. However, approximately 90% of the generated data is unstructured [8]. The multimedia data collected from the Internet and mobile devices is a typical example of this unstructured data [6]. Fully utilizing the collected data to generate commercial value is a huge challenge in many scenarios [9]. Many data preprocessing and machine learning methods are proposed to extract useful information from the data. Machine learning technology enables many aspects of modern people’s life, including web search, recommendation systems, natural language understanding as well as social networking [10–15]. Machine learning facilitates many services to Internet of Vehicles (IoV) [16]. With the aid of multiple sensors deployed in vehicles and roads, cloud platforms collect traffic information and current status from the vehicles [17]. Via data analysis and prediction, drivers are provided by accurate travel time estimation to improve their driving plans [18].