A novel dynamics reconstruction framework based on adversarial recurrent neural networks

The network assumes diverse roles in both topology and function. The topology of the network represents the connection relationship among nodes, while the functions within the network are typically manifested in the dynamic processes on the network. In this paper, we propose a network dynamics reconstruction framework that combines generative adversarial network (GAN) and recurrent neural networks (RNNs) to address the issue of network dynamics reconstruction with known network topology. In order to verify the feasibility and accuracy of the proposed method, reconstruction experiments of three different network dynamics are carried out in grid network, random network, small-world network and scale-free network. Compared with the traditional three recurrent neural networks, the experimental results show that the improved method proposed in this paper can reduce the reconstruction error by at least 7% and up to 82%, which fully proves the superiority of the method in this paper.