Clustered genetic search in continuous landscape exploration
Section snippets
Statement of the problem
There is a wide range of global optimization methods which are aimed at finding a single local minimizer (or maximizer) x+ of the real valued objective Φ in an admissible set . However, there is often a need of locating all available minimizers (maximizers) as well as detecting their basins of attraction (see definition below). One can mention inverse parametric problems being classical ambiguous problems with many minimizers, such as tasks appearing in the ultrasonographic defectoscopy for
Selected results of genetic algorithms asymptotic theory
This section is to present some theoretical results that will be used in Section 3 as a background for analysis of asymptotic properties of two clustered genetic search methods introduced there. In Section 2.1 we introduce some basic elements of Simple Genetic Algorithm (SGA) theory. We interpret SGA as a measure-transforming system and conclude with two theorems which are tools for asymptotic analysis of both clustered genetic search methods. Section 2.1 describes also comprehensively the idea
Two examples of clustered genetic search
For each strategy below we define the positive fitness function in the space of genetic codes (genetic universum), that expresses the source minimization problem for Φ on D. The encoding function represents the N-dimensional vector on D defined in the standard way (see eg. Goldberg, 1989).
Applications
In order to illustrate some properties of the presented methods, we show results of tests for two functions: two-dimensional sinus and a simple chemical interaction potential. The tests for the sinus function ((10) and Fig. 2a)illustrate the ability of both methods to recognize many basins in the case of a multimodal function (see Fig. 2, Fig. 3). The minima of the function constitute some one-dimensional manifolds, which are an additional difficulty.
However this
Concluding remarks
The combination of genetic algorithms and clustering methods in which the genetic sample is classified exhibits high positive synergy in solving global optimization problems. There are the following advantages of this approach:
- •
central parts of the basins can be detected and approximated;
- •
evolution landscape can be partially explored and recognized;
- •
there is a possibility to analyze the stability of minimizers;
- •
the number of local time-expansive searches can be delimit to one in each basin when the
References (23)
- et al.
Genetic clustering algorithms
European Journal of Operational Research
(2001) - et al.
Genetic algorithm-based clustering technique
Pattern Recognition
(2000) - et al.
A genetic clustering algorithm for data with non-spherical-shape clusters
Pattern Recognition
(2000) - et al.
A genetic approach to the automatic clustering problem
Pattern Recognition
(2001) Genetic clustering as a parallel algorithm for approximating basins of attraction
Lecture Notes in Computer Science
(2004)- Cabib, R., Schaefer, R., Telega, H., 1998. A Parallel Genetic Clustering for Inverse Problems. Lecture Notes in...
Finite Elements Method for Elliptic Problems
(1978)- et al.
Maximum likelihood from incomplete data via EM algorithm
Journal of the Royal Statistical Society Series B
(1977) - et al.
Pattern Classification and Scene Analysis
(1973) Genetic Algorithms in Search, Optimization and Machine Learning
(1989)
Cited by (15)
Region-based memetic algorithm with archive for multimodal optimisation
2016, Information SciencesHierarchic Genetic Strategy with maturing as a generic tool for Multiobjective Optimization
2016, Journal of Computational ScienceCitation Excerpt :Hierarchic strategies are also a subject of hybridization. They can be combined with local methods, e.g. local gradient-based search [30], and with clustering ([31] and [23], for HGS and MO-HGS, respectively). In order to outline the proposed MO-mHGS algorithm, it is necessary to define the input parameters which will be later referenced in the description: tm is the chosen metaepoch length, h specifies the maximum allowed height of the populations tree, ξmax is the maximum number of living children a tree node is allowed to have, and ξ defines sproutiveness – the amount of children created after each metaepoch.
Understanding measure-driven algorithms solving irreversibly ill-conditioned problems
2022, Natural ComputingAsynchronous actor-based approach to multiobjective hierarchical strategy
2020, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)Misfit landforms imposed by Ill-conditioned inverse parametric problems
2018, Computer ScienceTwo-phase strategy managing insensitivity in global optimization
2017, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)