Abstract We develop a deep-learning technique to infer the nonlinear velocity field from dark matter density field. The architecture we use is “U-net” style convolutional neural network, which consists of 15 convolution layers and 2 deconvolution layers. This setup maps three-dimensional 32 3 voxels or momentum fields 20 voxels. Through analysis simulation with resolution h ?1 Mpc, find that ne...