Special issue on trends & advances to mine intelligence from ambient data

Abstract

Ambient data is generated from a collective of smart devices , connected environments and pervasive platforms, so well-integrated and finely tuned that we no longer think about connecting, syncing, or controlling. The techniques and hardware involved in the development of Ambient Intelligence Systems are under a state of continuous evolution. This allows to develop hardware adapted systems and to deal with situations that were not approachable in different case studies. The use of new protocols, wireless sensor networks and IoT devices has allowed to incorporate new sources of ambient information; these can be processed thanks to the improvements in computing capacities and the reduced costs of the different devices. As a result, there is an increase in the use cases to which new trends and techniques in ambient analytics to obtaining ambient intelligence can be applied. This Special Issue focuses on the use of new trends in WSN, IoT, embedded systems and devices that incorporate Artificial Intelligence techniques or Distributed Artificial Intelligence (multi-agent systems, virtual organizations, classifiers, neural networks, Bayesian networks, etc.) for conducting ambient analytics. All the papers selected for this special issue have underwent two rounds of rigorous peer-review process. Based on the reviewers' feedback, as well as the evaluations of the Guest Co-editors, 18 papers were selected for this special issue. One of the papers is by Lafontaine et al. [1] entitled "An open vibration and pressure platform for fall prevention with a reinforcement learning agent". The authors of this paper propose an inexpensive open vibration platform equipped with pressure sensors. The platform is built from easily available electronic components to be used as a tool by physiotherapists in order to help them in their evaluation of the postural control of individuals at risk of postural imbalance. The platform has been built to be easily reproducible by the scientific community. Moreover, the computer code necessary to make it work is fully open source and can be used in any non-commercial applications. A first version of the platform was tested with 7 healthy human participants. A simple reinforcement learning agent was deployed and tested to automatically calibrate the vibration motors for optimal stimulation. The agent exploited computer vision to capture the data from a force platform commercially available and use it as ground truth. Also, a second version of the platform was built and presented. That version is have been validated clinically with both healthy and impaired human participants. Another paper is by Wang et al. [2] entitled "Revealing the hidden features in traffic prediction via entity embedding". This paper provides a NN-based model to predict traffic flow of a bike-sharing system in Suzhou, China. The model only uses external and discrete variables like weather, places of interest (POIs), and holiday periods. The authors applied both entities embedding and one-hot encoding for the data pre-processing of these variables. The results show that (1) Entity embedding can effectively increase the continuity of categorical variables and slightly improve the prediction efficiency for the NN model; and (2) The hidden relationship in variables can be identified through visual analysis, and the * Ansar-Ul-Haque Yasar