Narrow Band-Internet of Things (NB-IoT) is an emerging cellular-based radio access technology, which offers a range of flexible configurations for different coverage enhancement (CE) groups to provide reliable uplink connections for massive IoT devices with diverse data traffic. To optimize the number of served IoT devices, the uplink resource configurations need to be adjusted in real-time according to the dynamic traffic, this brings the challenge of how to select the configurations at the Evolved Node B (eNB) in the multiple CE groups scenario with high-dimension and interdependency. To tackle this challenge, multi-agent reinforcement learning (RL) is proposed as a promising solution, where the RL agent (i.e., implemented at the eNB) automatically updates the uplink resource configuration by interacting with the environment (i.e., the communication procedures in NB-IoT). In this talk, how the machine learning techniques such as deep learning, artificial neural networks (ANN) can be used dynamically to solve the numerous challenges in the Internet of Things (IoT) will be presented.