Editing Training (section)

=== Artificial-intelligence feedback ===

Learning processes developed for [[artificial intelligence]] are typically also known as training. [[Evolutionary algorithm]]s, including [[genetic programming]] and other methods of [[machine learning]], use a system of [[feedback]] based on "[[fitness function]]s" to allow [[computer program]]s to determine how well an entity performs a task. The methods construct a series of programs, known as a “population” of programs, and then automatically test them for "fitness", observing how well they perform the intended task. The system automatically generates new programs based on members of the population that perform the best. These new members replace programs that perform the worst. The procedure repeats until the achievement of optimum performance.<ref>
''Genetic Programming: An Introduction'', Wolfgang Banzhaf, [[Peter Nordin]], Robert E. Keller, and Frank D. Francone, Morgan Kaufmann Publishers, Inc., 1998
</ref>
In [[robotics]], such a system can continue to run in [[Real time (media)|real-time]] after initial training, allowing [[robot]]s to adapt to new situations and to changes in themselves, for example, due to wear or damage. Researchers have also developed robots that can appear to mimic simple human behavior as a starting point for training.<ref>
{{cite web  |url=http://www.etek.chalmers.se/~almir/  |title=HR-2 Robot can mimic simple human behavior  |url-status=dead  |archive-url=https://web.archive.org/web/20070607174912/http://www.etek.chalmers.se/~almir/  |archive-date=2007-06-07  }}</ref>