Ab initio nuclear theory aims to develop a predictive understanding of nuclei in terms of the nuclear forces between constituent nucleons. A major challenge is to solve the nuclear many-body problem, due to the complex many-body correlations induced by nuclear forces. In this talk, I will introduce a novel nuclear many-body method, the neural-network quantum Monte Carlo approach. This approach utilizes the strong expressive power of neural networks to encompass many-body correlations. Its accuracy can rival or even surpass that of conventional quantum Monte Carlo methods, which are based on imaginary-time propagation and suffer from the sign problem. I will also introduce its recent application in peripheral neutron-alpha scattering. This process is demonstrated to be a clean and sensitive probe to the long-range three-nucleon forces predicted by chiral symmetry.