Genetic Programming that Learns to Write Programs from Programming History
DescriptionGenetic Programming is a class of evolution inspired machine programming techniques. It has been applied to practical design problems, generating original and practical designs that are adopted by the industry. Two shortcomings of existing GP designs are identified: (1) suggested programs may repeat itself; a previously reported design may be re-generated. This is a great waste of computational resources, exacerbated by the fact that designs are computationally expensive and time consuming to evaluate. (2) There is no capability for machine learning. Past design failures should be learned so that design mistakes are not repeated but rather propose new, creative designs. This research aims to overcome the above two shortcomings by novel non-revisting genetic programming designs, as well as synergy of genetic programming and machine learning.
|Effective start/end date||1/05/10 → 15/10/12|