Training is a process to improve one's capacity or performance through the acquisition of knowledge or skills specific for the trained task. Although behavioral performance would be improved monotonically and reach a plateau as the learning progresses, neurophysiological signal shows different patterns like a U-shaped curve. One possible account for the phenomenon is that the brain first works hard to learn how to use task-relevant areas, followed by improvement in the efficiency derived from disuse of irrelevant brain areas for good task performance. Here, we hypothesize that topology of the brain network would show U-shaped changes during the training on a piloting task. To test this hypothesis, graph theoretical metrics quantifying global and local characteristics of the network were investigated. Our results demonstrated that global information transfer efficiency of the functional network in a high frequency band first decreased and then increased during the training while other measures such as local information transfer efficiency and small-worldness showed opposite patterns. Additionally, the centrality of nodes changed due to the training at frontal and temporal sites. Our results suggest network metrics can be used as biomarkers for quantifying the training progress, which can be differed depending on network efficiency of the brain.