%0 Journal Article %1 aler:2002:AI %A Aler, Ricardo %A Borrajo, Daniel %A Isasi, Pedro %D 2002 %J Artificial Intelligence %K Multi-strategy Planning Speedup algorithms, genetic learning, programming, %N 1-2 %P 29--56 %R doi:10.1016/S0004-3702(02)00246-1 %T Using genetic programming to learn and improve control knowledge %U http://citeseer.ist.psu.edu/511810.html %V 141 %X The purpose of this article is to present a multi-strategy approach to learn heuristics for planning. This multi-strategy system, called HAMLET-EVOCK, combines a learning algorithm specialised in planning () and a genetic programming (GP) based system (: Evolution of Control Knowledge). Both systems are able to learn heuristics for planning on their own, but both of them have weaknesses. Based on previous experience and some experiments performed in this article, it is hypothesised that handicaps are due to its example-driven operators and not having a way to evaluate the usefulness of its control knowledge. It is also hypothesized that even if control knowledge is sometimes incorrect, it might be easily correctable. For this purpose, a GP-based stage is added, because of its complementary biases: GP genetic operators are not example-driven and it can use a fitness function to evaluate control knowledge. and are combined by seeding initial population with control knowledge. It is also useful for to start from a knowledge-rich population instead of a random one. By adding the GP stage to , the number of solved problems increases from 58% to 85% in the blocks world and from 50% to 87% in the logistics domain (0% to 38% and 0% to 42% for the hardest instances of problems considered).

@article{aler:2002:AI, abstract = {The purpose of this article is to present a multi-strategy approach to learn heuristics for planning. This multi-strategy system, called HAMLET-EVOCK, combines a learning algorithm specialised in planning () and a genetic programming (GP) based system (: Evolution of Control Knowledge). Both systems are able to learn heuristics for planning on their own, but both of them have weaknesses. Based on previous experience and some experiments performed in this article, it is hypothesised that handicaps are due to its example-driven operators and not having a way to evaluate the usefulness of its control knowledge. It is also hypothesized that even if control knowledge is sometimes incorrect, it might be easily correctable. For this purpose, a GP-based stage is added, because of its complementary biases: GP genetic operators are not example-driven and it can use a fitness function to evaluate control knowledge. and are combined by seeding initial population with control knowledge. It is also useful for to start from a knowledge-rich population instead of a random one. By adding the GP stage to , the number of solved problems increases from 58% to 85% in the blocks world and from 50% to 87% in the logistics domain (0% to 38% and 0% to 42% for the hardest instances of problems considered).}, added-at = {2008-06-19T17:35:00.000+0200}, author = {Aler, Ricardo and Borrajo, Daniel and Isasi, Pedro}, biburl = {https://www.bibsonomy.org/bibtex/23647840dc5672d62fd6503d0397f28cd/brazovayeye}, doi = {doi:10.1016/S0004-3702(02)00246-1}, interhash = {7052414a1f290a67b9ee3351725c62e4}, intrahash = {3647840dc5672d62fd6503d0397f28cd}, journal = {Artificial Intelligence}, keywords = {Multi-strategy Planning Speedup algorithms, genetic learning, programming,}, month = {October}, notes = {Hamlet, EvoCK, PRODIGY 4.0}, number = {1-2}, pages = {29--56}, timestamp = {2008-06-19T17:35:31.000+0200}, title = {Using genetic programming to learn and improve control knowledge}, url = {http://citeseer.ist.psu.edu/511810.html}, volume = 141, year = 2002 }