Video 13

GNN trained on a reaction-diffusion simulation based on the "Rock-Paper-Scissor" automaton over a mesh of 10^4 3D vector nodes (4,000 time-steps). The amplitudes of the vector components are represented as RGB colors. Vector nodes have varying diffusion coefficients that modulate the reaction. (Left) Validation dataset with different initial conditions than the training dataset. (Right) Rollout inference of the trained GNN.