Neuroevolution
|
|
Neuroevolution, or neuro-evolution, is the use of genetic algorithms to train artificial neural networks. It is useful for applications such as games and robotic motor control, where it is easy to measure a network's performance at a task but difficult or impossible to create a syllabus of correct input-output pairs for use with a supervised learning algorithm.
There are many different neuroevolutionary algorithms. A distinction is made between those that only evolve the values of the connection weights for a network of pre-specified topology, vs. those that evolve the topology of the network in addition to the weights. (However, there are no standardized terms for the distinction.) Networks that evolve both the connection weights and the topology are sometimes called TWEANNs (Topology & Weight Evolving Artificial Neural Networks).
Direct-encoding methods use floating-point numbers in the genetic algorithm's chromosomes to directly specify the values of a network's connection weights. More sophisticated indirect encoding methods are also possible.
See also
External links
- University of Texas neuroevolution page (http://nn.cs.utexas.edu/area-view.php?RECORD_KEY(Areas)=AreaID&AreaID(Areas)=7) (has downloadable papers on methods and applications)