Manuscript Summary - Neuroevolutionary learning of particles and protocols for self-assembly
This is a simple simulated model of self-assembly on a surface, where we have a neural network which:
1) decides how to change the growth conditions as the nano-structure develops based on its current form. This is a feedback and control scheme which is evolved to either promote a particular target structure, or, alternatively, seeks out new and unseen structures on its own (by growing new targets which are maximally distant from previously grown ones).
2) decides the types of interactions between the building blacks. These kind of design decisions would be realizable in the lab through the choice of functionalization / linker groups, a particular DNA sequence (for self-assembled DNA structures) etc.
The fact that we use a physical simulator means every single design and structure discovered is makeable by definition. No special processes required.