In this project, new machine learning (ML)-assisted algorithms will be designed to estimate and to learn the multipath network by extending known super-resolution techniques for sparsity exploitation. These algorithms will take advantage of the fixed locations of base stations and infrastructure around them, and hence of repeated patterns of mobile terminals along sidewalks and roads, and in buildings, etc. These dynamic patterns will be learned at a high resolution, hence help obviating the need for expensive feedback of channel state information. The project will also address a rich set of important uplink and downlink system design problems, such as joint communication, coding, and networking optimizations. These problems are generated due to the new multipath network paradigm. The goal would be to set up and to maintain communication through many almost interference-free paths (thin as pins), thereby resulting in highly effective utilization and sharing of spectrum via a dense reuse of frequency across space, and the precise avoidance of interference to the primary users of that spectrum.