Evolutionary systems & genetic algorithms

KStanley is not an artist, but his development of the NeuroEvolution of Augmenting Topologies (NEAT) method has been applied to image making by his students. It has also been applied to video games, the control of animation particle systems, and recently the evolutionary training of virtual dancers. The NEAT method evolves a neural network via complexification. I.e. the system starts with a very simple network and then slowly adds more nodes, connections, and weights as it gets closer and closer to the desired solution.

KStanley is not an artist, but his development of the NeuroEvolution of Augmenting Topologies (NEAT) method has been applied to image making by his students. It has also been applied to video games, the control of animation particle systems, and recently the evolutionary training of virtual dancers. The NEAT method evolves a neural network via complexification. I.e. the system starts with a very simple network and then slowly adds more nodes, connections, and weights as it gets closer and closer to the desired solution.

The student evolutionary art project is available as a web-based Java application [http://picbreeder.org/ here]. A website where you can download the video game developed using the NEAT method is available [http://nerogame.org/ here]. Stanley's Ph.D. dissertation which established the NEAT method is available [http://nn.cs.utexas.edu/downloads/papers/stanley.phd04.pdf here], and his web page is [http://www.cs.ucf.edu/~kstanley/ here]. This relatively recent work holds much promise to address the long standing challenge of effectively designing neural networks using evolutionary methods.

The student evolutionary art project is available as a web-based Java application [http://picbreeder.org/ here]. A website where you can download the video game developed using the NEAT method is available [http://nerogame.org/ here]. Stanley's Ph.D. dissertation which established the NEAT method is available [http://nn.cs.utexas.edu/downloads/papers/stanley.phd04.pdf here], and his web page is [http://www.cs.ucf.edu/~kstanley/ here]. This relatively recent work holds much promise to address the long standing challenge of effectively designing neural networks using evolutionary methods.