TLDR: Imagine you have a bridge with well known stress characteristics, loads, etc. Now you want to run an optimizer to produce a lightweight bridge made out o

TLDR: Imagine you have a bridge with well known stress characteristics, loads, etc. Now you want to run an optimizer to produce a lightweight bridge made out of a material that uses repeating structures throughout. This exists today but takes a long time to run, you can get all sorts of interesting material compositions out of such programs. Instead of doing the above train a deep neural network to predict from a set of known stress properties of shapes made out of different types of materials to a set of the structures of those materials. (Massive over simplification: should I use pyramids or cubes as my basic building block for a material?) After training on millions of known examples, the network is able to take in a desired set of stress characteristics and produce the material which will have those characteristics, very quickly. They seem to be particularly interested in predicting materials to make up bone replacements.