Deep Learning with Partially Labeled Data for Radio Map Reconstruction

Retrieving the exact position of the connected objects has become an important feature of the Internet of Things (IoT). Such connected objects have indeed been widespread over the last few years thanks to the low cost of the radio integrated chips and sensors and their possibility of being embedded in plurality of the devices. By this they can help in fast development of large-scale physical monitoring and crowdsensing systems (like smart cities, factories, transportation, etc.). For the location-dependent application and services these abilities to associate accurate location with physical data gives huge opportunities [25]. For example, the fine-grain and dynamic update of air pollution and/or weather maps could benefit from geo-referenced mobile sensing [1] (e.g., aboard taxis, buses, bicycles...), thus continuously complementing the data from static stations. One of the localization techniques is Global Positioning System (GPS) which has been widely used over the past decades.