#compositional pattern producing networks

[procedural-generation]

EndlessForms

EndlessForms is a site by the Creative Machines Laboratory at Cornell that uses evolutionary developmental processes to generate voxel objects based on the genes’ relative geometric position. The site explores the generative space that emerges from Compositional Pattern Producing Networks. Unlike other ways of simulating growth, CPPNs are locally independent: each individual component grows without knowing what its neighbors are doing.

One reason that CPPNs are useful for procedural generation is that they can effectively represent a highly complex generative space because they can vary both the topology and activation functions of the network.

http://endlessforms.com/