Advertisement
Published in:
2008 | OriginalPaper | Chapter
Evolution of Descent Directions
Authors : Alejandro Sierra Urrecho, Iván Santibáñez Koref
Published in: Adaptive and Multilevel Metaheuristics
Publisher: Springer Berlin Heidelberg
Activate our intelligent search to find suitable subject content or patents.
Select sections of text to find matching patents with Artificial Intelligence. powered by
Select sections of text to find additional relevant content using AI-assisted search. powered by
Estimation of distribution algorithms proceed by sampling new solutions from a probability distribution learnt in an evolutionary way. This involves keeping track of a population of candidate solutions and updating distribution parameters from the best of these candidates. We propose to substitute this population of solutions by one of descent directions. New solutions will no longer be sampled but interpolated along each direction in a deterministic way. Even when strong correlations between dimensions are present, sampling new directions from a product of independent one-dimensional Gaussian distributions is enough because covariances are captured by the directions. Despite its simplicity, our algorithm can address problems such as the rotated cigar function with state of the art performance and without any covariance calculation.