Alogorithms like what you are describing (minus the learning part) are a very active area of research, its quite exciting! https://en.wikipedia.org/wiki/Topology_optimization