Abstract | ||
---|---|---|
This paper investigates the search efficiency of a class of adaptive memetic algorithms where the pivot function, depth, and definition of the local search operators are co-evolved alongside a population of potential solutions to the problem in hand. Such co-evolutionary mechanism requires a means for assigning meme fitness based in some way on the improvement they cause in solutions at a particular stage in the search process. We examine schemes based on both the extremal and mean improvement caused, and compare these to the implicit self-adaptive scheme. Simultaneously we examine the effect of using different fixed or adaptive pivot functions and depths of search. Results show that provided the fitness is correctly assigned the system successfully adapts the global/local search trade-off via evolution of the memes' search depth. The system is also able to adapt the optimal choice of greedy or steepest ascent. Unlike recent work on adaptive operator choice, results suggest that a fitness based on a meme's mean, rather than extremal affect provides more reliably effective optimisation results. Despite the close coupling between the two population, the self-adaptive schemes which use implicit fitness assignment are less successful than a well designed co-evolutionary scheme. Finally we examine the effect of changing the size of the meme pool and show that a surprisingly large number can be processed and benefit evolution. |
Year | DOI | Venue |
---|---|---|
2010 | 10.1109/CEC.2010.5586401 | IEEE Congress on Evolutionary Computation |
Keywords | Field | DocType |
evolutionary computation,search problems,adaptive operator choice,adaptive pivot functions,coevolutionary memetic algorithms,implicit self-adaptive scheme,local search operators,meme fitness,memepool sizes,search efficiency | Memetic algorithm,Population,Mathematical optimization,Algorithm design,Computer science,Close coupling,Evolutionary computation,Operator (computer programming),Artificial intelligence,Local search (optimization),Memetics,Machine learning | Conference |
ISBN | Citations | PageRank |
978-1-4244-6909-3 | 2 | 0.36 |
References | Authors | |
18 | 1 |