Neural Relational Inference with Efficient Message Passing Mechanisms

Many complex processes can be viewed as dynamical systems of interacting agents. In many cases, only the state sequences individual agents are observed, while relations and rules unknown. The neural relational inference (NRI) model adopts graph networks that pass messages over a latent to jointly learn dynamics based on observed data. However, NRI infers independently suffers from error accumulation in multi-step prediction at learning procedure. Besides, relation reconstruction without prior knowledge becomes more difficult systems. This paper introduces efficient message passing mechanisms with structural address these problems. A interaction mechanism is proposed capture coexistence all relations, spatio-temporal use historical information alleviate accumulation. Additionally, knowledge, symmetry special case, introduced for better experimental results simulated physics show method outperforms existing state-of-the-art methods.