Researchers affiliated with Google and the Paradigms of Intelligence Team explore the concept of reverse-engineering complex patterns from simple rules using cellular automata. They aim to create systems that can learn local rules to produce desired patterns, blending Neural Cellular Automata (NCA) with Differentiable Logic Gates Networks. The NCA can learn patterns and behaviors but struggle with discrete states, while DLGNs discover logic circuits but lack recurrent settings. By combining these two concepts, the team hopes to pave the way for programmable matter – Computronium. They conduct experiments to showcase the learning capabilities and effectiveness of these new systems.
https://google-research.github.io/self-organising-systems/difflogic-ca/?hn