HyperNEAT

Hypercube-based NEAT, or HyperNEAT,^[1] is a generative encoding that evolves artificial neural networks (ANNs) with the principles of the widely used NeuroEvolution of Augmented Topologies (NEAT) algorithm developed by Kenneth Stanley.^[2] It is a novel technique for evolving large-scale neural networks using the geometric regularities of the task domain. It uses Compositional Pattern Producing Networks ^[3] (CPPNs), which are used to generate the images for Picbreeder.org Archived 2011-07-25 at the Wayback Machine and shapes for EndlessForms.com Archived 2018-11-14 at the Wayback Machine. HyperNEAT has recently been extended to also evolve plastic ANNs ^[4] and to evolve the location of every neuron in the network.^[5]

Querying the CPPN to determine the connection weight between two neurons as a function of their position in space. Note sometimes the distance between them is also passed as an argument.

Applications to date

• Multi-agent learning^[6]
• Checkers board evaluation^[7]
• Controlling Legged Robots^{[8][9][10][11][12][13]}video
• Comparing Generative vs. Direct Encodings^[14][15][16]
• Investigating the Evolution of Modular Neural Networks^[17][18][19]
• Evolving Objects that can be 3D-printed^[20]
• Evolving the Neural Geometry and Plasticity of an ANN^[21]