Randomness improves learning and stability in neural cellular automata
Where Does Randomness Matter in Neural Cellular Automata?
Machine LearningNeural and Evolutionary Computing
Summary
Neural cellular automata (NCA) are computer models that update many small parts, or cells, to grow patterns or shapes. This study shows that random updates during the learning process help the models learn better rules and can keep the final pattern stable even when randomness is removed later. The researchers also found that different kinds of evaluation and damage recovery affect how well the model retains the pattern over time. This helps separate the roles of randomness in training from randomness in execution.
What this means in practice
- •For machine learning engineers: Design more reliable neural cellular automata by using random asynchronous updates during training to ensure stable pattern growth under deterministic execution.
- •For computational biologists: Model tissue growth and regeneration processes with NCAs that distinguish learning stability from execution behavior for better simulation accuracy.
Authors
Fei Zuo, Jiaqi Shi, Yujing Liu
Abstract
Stochastic cell updates are often used throughout the life of a neural cellular automaton (NCA), from backpropagation through time to final rollout. This leaves two questions entangled: does update randomness help learn a useful rule, and must that randomness remain at execution? We separate training and evaluation update modes in controlled Growing NCA experiments, then vary the states shown during training. Under the standard constant-rate persist recipe, asynchronous training passes the short-horizon quality test in 10/10 runs, compared with 3/10 synchronous runs. All ten asynchronous models also retain the target for 4,096 steps under deterministic evaluation. For a scalar translation-invariant lattice, we derive an exact mean-square criterion: random masking can damp mean modes, but it also injects variance, and a mean-only test misclassifies four non-marginal settings. Finally, among 30 models that all pass the same reconstruction test, eight of ten grow-trained models become off-target at 4,096 steps, while all persist and regenerate models retain the target; damage recovery separates persist from regenerate. The results distinguish optimization reliability, execution mode, and task-specific behavior instead of treating them as one stability property.