Growing Neural Cellular Automata

This introduces research that implements the process of morphogenesis in living organisms as a differentiable model using Neural Cellular Automata (NCA). It explores the phenomenon of self-organization, in which cells communicate with their neighbors to autonomously form complex structures.

Models are classified into three types according to their training objective:

• Growing models: Aim to generate patterns, but have a low ability to persistently maintain the pattern.
• Persistent models: Aim for the long-term maintenance of patterns, and develop regeneration ability during training even without explicit instruction.
• Regenerating models: Models in which the ability to repair damaged areas is further strengthened by inflicting damage on the pattern during training.

This research explores the algorithmic principles behind how complex, robust biological structures emerge from simple rules, and how a population of cells can perceive and regulate the overall form.