Many resource optimization problems in large and complex networked systems are NP-hard. However, the underlying network may changes quite swiftly, in terms of the channel conditions, demand patterns, and even topology. As a result, it is extremely challenging to design fast and efficient solutions that can keep up with the changing network dynamics. On the other hand, modern wireless and wireline networks have already collected a large volume of operational data. We plan to develop a data-guided operational framework that can transform such “data” to improvements in key network functions. Specifically, large volumes of historical data can be processed offline to learn representative features and to pre-compute a suitable set of resource management decisions. Then, combined with finer and lower-complexity online dynamic adaption based on instantaneous network conditions and user profiles, this data-guided approach can more effectively manage the computational complexity, and thus achieve higher efficiency and adaptivity.