While the emerging mobile cloud technology provides a promising solution for mining big multimedia data in wireless networks, there exist several limitations imposed by the energy-constrained mobile devices, irresponsive delivery of mined stream, and huge drain on server-side resources. This paper proposes a mobile cloud-based stream mining system that addresses three key metrics: energy consumption of the cloud, accuracy of mining result and responsiveness, as well as the energy consumption of battery-powered mobile users. Since optimizing all metrics simultaneously is impossible in practice, the cloud operator needs to properly tradeoff these metrics. However, this is not straight forward as the environment in which the system operates randomly changes over time, and the distribution of the underlying stochastic process is often unknown a priori. To address these issues, the authors leverage state-of-the-art control techniques and develop an online computational resource provisioning and transmission rate selection algorithm to minimize the classification-energy cost. Each mobile user can thus decide its transmission rate, while the cloud operator decides its computational resource provisioning for stream mining. The distinguishing feature of online execution makes the proposed solution an appealing candidate for future mobile cloud supporting interactive stream mining and for realizing the full potential of big data in wireless networks.
This paper has also been recommended as a "Distinguished Paper" in the IEEE ComSoc MMTC Reviewer Letter in February 2014.