Representation by neural networks, in particular by restricted Boltzmann machines (RBM), has provided a powerful computational tool to solve quantum many-body problems. An important open question is how to characterize which class of quantum states can be efficiently represented with RBMs. Here, we show that RBMs can efficiently represent a wide class of many-body entangled states with rich exotic topological orders. This includes: (1) ground states of double semion and twisted quantum double models with intrinsic topological orders; (2) states of the AKLT model and two-dimensional CZX model with symmetry protected topological orders; (3) states of Haah code model with fracton topological order; (4) (generalized) stabilizer states and hypergraph states that are important for quantum information protocols. One twisted quantum double model state considered here harbors non-abelian anyon excitations. Our result shows that it is possible to study a variety of quantum models with exotic topological orders and rich physics using the RBM computational toolbox.Introduction.-Deep learning has become a powerful tool with wide applications [1,2]. Recently, deep learning methods have attracted considerable attention in quantum physics [3,4], especially for attacking quantum many-body problems. The difficulty of quantum manybody problems mainly originates from the exponential growth of the Hilbert space dimension. To overcome this exponential difficulty, researchers traditionally use tensor network methods [5][6][7] and Quantum Monte Carlo (QMC) simulation [8]. However, QMC methods suffer from the sign problem [9]; Tensor network methods have difficulty to deal with high dimensional systems [10] or systems with massive entanglement [11]. These issues call for mew method.