Title
Semantic building blocks in genetic programming
Abstract
We present a new mechanism for studying the impact of subtree crossover in terms of semantic building blocks. This approach allows us to completely and compactly describe the semantic action of crossover, and provide insight into what does (or doesn't) make crossover effective. Our results make it clear that a very high proportion of crossover events (typically over 75% in our experiments) are guaranteed to perform no immediately useful search in the semantic space. Our findings also indicate a strong correlation between lack of progress and high proportions of fixed contexts. These results then suggest several new, theoretically grounded, research areas.
Year
DOI
Venue
2008
10.1007/978-3-540-78671-9_12
EuroGP
Keywords
Field
DocType
semantic building block,crossover event,high proportion,semantic space,new mechanism,semantic action,genetic programming,research area,strong correlation,fixed context,subtree crossover
Boolean function,Crossover,Computer science,Tree (data structure),Evolutionary computation,Genetic programming,Theoretical computer science,Correlation,Artificial intelligence,Machine learning,Semantic computing,Semantic space
Conference
Volume
ISSN
ISBN
4971
0302-9743
3-540-78670-8
Citations 
PageRank 
References 
69
3.47
13
Authors
3
Name
Order
Citations
PageRank
Nicholas Freitag McPhee140432.94
Brian Ohs2693.47
Tyler Hutchison3955.54