An Approach for Estimating Separability and its Application on High Dimensional Optimization

作者：Landa Ricardo^*; Rojas Yazmin; Toscano Pulido Gregorio

来源：14th International Conference on Genetic and Evolutionary Computation Conference (GECCO), 2012-07-07 To 2012-07-11.

DOI：10.1145/2330163.2330295

• Rectangle condition
• correlation coefficients
• differential evolution
• large scale optimization
• non-separable problems

In this paper, we propose an approach for measuring the level of separability (in a relaxed sense) among variables, making use of the rectangle condition for separable functions. This approach is then used in a differential evolution-based algorithm for high dimensional optimization. The decision variables are associated into groups by their estimated level of separability. Such estimation is refined throughout generations, depending on the area being currently explored. Results are shown from 50 to 10,000 variables. The experiments are performed with unimodal, multimodal, separable and non-separable functions. Comparison are shown with differential evolution alone, and with other algorithms of the state of the art.

• 出版日期2012