NeuroEvolution of Augmented Topologies
NeuroEvolution of Augmenting Topologies (NEAT) is the name of a genetic algorithm that evolves artificial neural networks . It was developed in 2002 by Ken Stanley at the University of Texas at Austin . Due to its practical applicability, the algorithm is used in various areas of machine learning . Both the topology and the weights of the connections in the neural network are evolved. The main characteristics of NEAT are:
- the assignment of an identification number ( innovation number ), which allows the advantageous recombination of different topologies,
- the recess formation by limiting recombination on a degree of relationship and
- the increasing diversity of the population with initial uniformity.
The HyperNEAT extension enables the evolution of significantly larger networks by utilizing the geometric structures of the given problem (e.g. the control of several legs).
Web links
- Original publication by Kenneth O. Stanley and Risto Miikkulainen: Evolving Neural Networks through Augmenting Topologies. (PDF; 456 kB) In: Evolutionary Computation. 10 (2) 2002, pp. 99-127.