We model wireless networks via Markov Decision Processing modeling which induces directed graph descriptions for the network state. These graphs are logical in that each node in the graph represents one state of the entire network. Traditional graph signal transforms exploit low pass features of graph signals to determine low dimensional representations of graph signals resulting in good graph signal approximation, sampling and interpolation strategies. It is becoming increasingly clear that such strategies do not map well to graph reductions for policy optimization in wireless networks. We will review our previous endeavors in this arena and discuss new findings which are promising with respect to the design of optimal policies for large scale wireless networks.