| Summary: | 碩士 === 國立中正大學 === 資訊工程研究所 === 103 === Evolutionary algorithms (EAs) using surrogate models are well known as surrogate-
assisted EAs, or meta-model based EAs. Surrogates is an ecient means of handling com-
plicated applications since the cost of tness evaluation can be reduced. There are many
powerful machine learning algorithms have been proved to be useful for modeling surrogates.
Recently, a novel eective learning method called sparse learning (or sparse dictionary learn-
ing) is proposed and widely applied. Sparse learning is successful in signal processing and
machine learning. The advantages of sparse learning included dimension reduction, feature
extraction, and robustness of noise, which are attractive functionality for surrogates. Based
on the theory of sparse learning, this study proposes using sparse learning as an surrogate
modeling method for the advantages described above. The experimental results show that
the proposed method not only perform well in the early stage of evolution but also obtain
acceptable solution quality when evolution terminated.
|
|---|
