Exploring Deep Learning for Efficient and Reliable Mobile Sensing

The nine articles in this special section focus on the use of deep learning program for mobile sensing technologies and applications. Smart devices equipped with a rich set of sensors, such as smartphones and wearables, are enabling an ever-increasing number of mobile applications, ranging from user behavior perception such as gesture recognition and health monitoring to context inference such as driving detection and hazard warning. However, reliably inferring high-value user behavior and context information from noisy and highly heterogeneous sensor data using mobile devices remains an open issue. To combat against the constraints of a mobile device, one conventional approach is to upload raw sensor data to the cloud and run complex inference algorithms remotely. This leads to inefficiency in terms of network overhead and large latency for mobile applications. Recently, deep learning techniques have made significant breakthroughs in achieving state-of-art inference performance in a variety of applications such as computer vision and natural language processing. In general, data modeling challenges found in mobile sensing can often be seen in deep-learningbased inference.