Learning-assisted evolutionary search for scalable function optimization: LEM(ID3)

Inspired originally by the Learnable Evolution Model(LEM) [5], we investigate LEM(ID3), a hybrid of evolutionary search with ID3 decision tree learning. LEM(ID3) involves interleaved periods of learning and evolution, adopting the decision tree construction algorithm ID3 as the learning method, and a steady state EA as the evolution component. In the learning periods, ID3 is used to infer rules that attempt to identify `good' regions for genes, based on the values of one or more other genes. The rules are then used to guide the generation of new individuals. Without any preliminary parameter tuning, we evaluate LEM(ID3) on the test suite of 25 functions designed the CEC 2005 special session on Real-Parameter Optimization. We describe the results, and in particular compare with the three most successful algorithms from the CEC 2005 competition; Sinha et al's K-PCX, and two versions of Auger and Hansen's CMA-ES. We find that LEM(ID3)'s performance is competitive with ...

• Type: Conference
• Year: 2010
• Where: CEC
• Authors: Guleng Sheri, David Corne
• Keywords: Artificial Intelligence, CEC 2010, Decision Tree, Decision Tree Learning, Involves Interleaved Periods