Yao, Liu, Liang, and Lin propose and test several new evolutionary algorithms that outperform existing ones on their set of benchmark test functions. The benchmark functions are detailed in an appendix.
A new, fast evolutionary programming (FEP) method, which performs better due to the Cauchy mutation higher probability of making longer jumps, is tested on 23 benchmark functions, including unimodal and multimodal functions.
The authors then propose an improved FEP (IFEP) method. IFEP is based on mixing different search biases of Cauchy and Gaussian mutations, which are experimentally tested and found to yield better results than either FEP or classical evolutionary programming (CEP) for most of the test functions.
Behavior of FEP versus CEP, which uses Gaussian mutation, is then assessed, based both on analysis and on empirical evidence.
A hybrid algorithm, based on landscape approximation (which transforms a complex problem into a simpler one without changing the optima to be found) and local search (LALS), is presented and experimentally tested on a subset of the problem collection. A comparison between the IFEB and LALS shows LALS has better global reliability, but is more time consuming than IFEP.
Finally, an evolutionary algorithm with N-dimensional approximation (EANA) and limited local search is tested on a subset of benchmark functions. This method appears to yield promising results.
It will be interesting to know how the new results in evolutionary optimization perform on real life problems where dimensionality is higher than the benchmark functions.