Learning beyond local view: Value and information in the bits

Given certain amount of resources available for acquiring network-state information, what should be learned? In this paper, we study this fundamental question for a Z channel where each user has certain local view and beyond that it is allowed to learn k-bits of global network state information. We show that if the interference is unknown to both the transmitters, the best learning strategy is to quantize the signal to interference ratio and reveal it to both transmitters. However, if the interference is known to at least one of the transmitters, then a two-dimensional quantization of the global channel state is the optimal utilization of the k-bits.