A CMA-ES for Mixed-Integer Nonlinear Optimization
Abstract
We propose a modification of CMA-ES for the application to mixed-integer problems. The modification is comparatively small. First, integer variables with too a small variation undergo an additional integer mutation. This mutation is also used for updating the distribution mean but disregarded in the update of covariance matrix and step-size. Second, integer variables with too a small variation are disregarded in the global step-size update alltogether. This prevents random fluctuations of the step-size.
Domains
Machine Learning [cs.LG]Origin | Files produced by the author(s) |
---|
Loading...