In mobile edge computing (MEC) networks, by offloading tasks (partially or completely) to the MEC server, it becomes possible to complete computation-intensive and latency-critical applications without communicating with the cloud center, resulting in dramatic reduction both in latency and energy consumption. Performance improvements depend on the offloading decisions at the user equipments (UEs) and computational resource allocation at the MEC server. In this paper, we aim to optimize the UE offloading data ratios and MEC computational resource allocation under delay constraints with the goal to minimize the global energy consumption. Both conventional optimization method and learning-based approach are studied. Simulation results are provided to compare the performances of different schemes.