Inspired by residual neural networks, the update rule outputs an incremental update to the cell’s state, which applied to the cell before the next time step. Finally, as the model becomes more robust at going from a seed state to the target state, the samples in the pool reflect this and are more likely to be very close to the target pattern, allowing the training to refine these almost completed patterns further. Since we trained our coupled system of cells to generate an attractor towards a target shape from a single cell, it was likely that these systems, once damaged, would generalize towards non-self-destructive reactions.