Agile Deep Learning UAVs Operating in Smart Spaces : Collective Intelligence Versus “Mission-Impossible”
Cochez, M., Periaux, J., Terziyan, V., & Tuovinen, T. (2018). Agile Deep Learning UAVs Operating in Smart Spaces : Collective Intelligence Versus “Mission-Impossible”. In P. Diez, P. Neittaanmäki, J. Periaux, T. Tuovinen, & O. Bräysy (Eds.), Computational Methods and Models for Transport: New Challenges for the Greening of Transport (pp. 31-53). Computational Methods in Applied Sciences, 45. Springer. doi:10.1007/978-3-319-54490-8_3
Published inComputational Methods in Applied Sciences
© Springer International Publishing AG 2018.
The environments, in which we all live, are known to be complex and unpredictable. The complete discovery of these environments aiming to take full control over them is a “mission-impossible”, however, still in our common agenda. People intend to make their living spaces smarter utilizing innovations from the Internet of Things and Artificial Intelligence. Unmanned aerial vehicles (UAVs) as very dynamic, autonomous and intelligent things capable to discover and control large areas are becoming important “inhabitants” within existing and future smart cities. Our concern in this paper is to challenge the potential of UAVs in situations, which are evolving fast in a way unseen before, e.g., emergency situations. To address such challenges, UAVs have to be “intelligent” enough to be capable to autonomously and in near real-time evaluate the situation and its dynamics. Then, they have to discover their own missions and set-up suitable own configurations to perform it. This configuration is the result of flexible plans which are created in mutual collaboration. Finally, the UAVs execute the plans and learn from the new experiences for future reuse. However, if to take into account also the Big Data challenge, which is naturally associated with the smart cities, UAVs must be also “wise” in a sense that the process of making autonomous and responsible real-time decisions must include continuous search for a compromise between efficiency (acceptable time frame to get the decision and reasonable resources spent for that) and effectiveness (processing as much of important input information as possible and to improve the quality of the decisions). To address such a “skill” we propose to perform the required computations using Cloud Computing enhanced with Semantic Web technologies and potential tools (“agile” deep learning) for compromising, such as, e.g., focusing, filtering, forgetting, contextualizing, compressing and connecting. ...