Although the IoT application use cases, presented herein with their DL-based implementations, demonstrate the potential of DL in IoT, there are still some open research challenges in many directions. In particular, research and development support are needed in many areas. A few of the key areas of research and development are dataset preprocessing, secure, and privacy-aware DL, aspects of handling big data, and resource-efficient training and learning. In the following sections, we briefly present these remaining challenges from a machine learning perspective, as well as from the perspective of IoT devices, edge/fog computing, and the cloud.





















































