Learning Autonomy in Management of Wireless Random Networks

This paper presents a machine learning strategy that tackles distributed optimization task in wireless network with an arbitrary number of randomly interconnected nodes. Individual nodes decide their optimal states coordination among other through varying backhaul links. poses technical challenge universal policy robust to random topology the network, which has not been properly addressed by conventional deep neural networks (DNNs) rigid structural configurations. We develop flexible DNN formalism termed message-passing (DMPNN) forward and backward computations independent topology. A key enabler this approach is iterative message-sharing arbitrarily connected The DMPNN provides convergent solution for numerous interactions. investigated various configurations power control networks, intensive numerical results prove its universality viability over approaches.