Dynamic Neural Cellular Automata

Neural cellular automata are a class of models capable of learning local interactions that give a desired emergent behaviour. Originaly developed in the context of growing images from single pixels, we extend this to learn dynamics from time series of images, and from Turing patterning in reaction-diffusion PDEs: https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1011589.