Cellular automata as convolutional neural networks by Gilpin, W. (2018)

tags

Cellular automata as CNNs

source

(Gilpin 2018)

Summary

This is one of the only attempt to represent a CA rule as a CNN I have come across. The author uses a deep CNN to learn a rule and studies various information-theoretic quantities in the activation patterns to evaluate the complexity of the rules.

Comments

I am personally very interested by the paper since it is an interesting direction for creating neural-network based rules that can be sampled and efficiently stored and applied. This work is also cited in (Mordvintsev et al. 2020).

Bibliography

1. William Gilpin. September 9, 2018. "Cellular Automata as Convolutional Neural Networks". http://arxiv.org/abs/1809.02942.
2. Alexander Mordvintsev, Ettore Randazzo, Eyvind Niklasson, Michael Levin. February 11, 2020. Distill 5 (2):e23. DOI. See notes