Finding the global optimum for a multimodal function has always been difficult. Evolutionary algorithms (EAs) have been used widely in tackling various multimodal functions with many local optima, with some success. Most of the work in EAs has concentrated on how to design better search operators or better representations for a difficult problem. A different approach is adopted in our study: rather than modifying an EA to solve a difficult problem, we approximate a difficult problem by a simpler one so that it is easier to solve. In particular, we try to approximate a hilly multimodal function by a smoother function so that it can be tackled by a hybrid local and evolutionary algorithm. Some preliminary experimental results will be presented in the talk.

The work is still in progress. I'd appreciate any feedback and comments.