Big data is set to offer new insights of many research areas, but a major challenge is the computation and storage cost in analyzing massive data. In this talk, we aim to embed computational thinking into traditional uncertainty quantification and develop a new direction of computationally efficient inference (COFFEE). Particularly, we propose a new nonparametric global testing based on random projection, which demonstrates superiority in reducing computational burden. Besides introducing the testing method, our theoretical contribution is to characterize the minimal computational cost that is needed to achieve the minimax optimal testing power.