Nanotechnology in Regenerative Medicine Methods and Protocols
Nanotechnology plays a key leading role in developing tools able to identify, measure, and study cellular events at the nanometric level as well as in contributing to the disclosure of unknown biological interactions and mechanisms, which opens the door f
- PDF / 243,001 Bytes
- 9 Pages / 430 x 660 pts Page_size
- 111 Downloads / 204 Views
1 Introduction In this paper, we propose a new population-based framework for combining local search (Brent Direct Search [1, 2]) with global exploration to solve single-objective unconstrained numerical optimization problems. The idea is to use knowledge about local optima found during the search to a) locate promising regions in the search space and b) identify the suitable step size to move from one optimum to others in each region. The general properties of the framework could be described as follow: • A population of individuals is maintained. Each individual is a mobile agent equipped with two heuristics: one internal local search to find local optima and one adaptive move length to get out of the current basin of attraction. • To successfully get out of the current basin as well as to approach promising areas, individuals need to share their knowledge about the search space with others. They do that by contributing their knowledge to a shared source called belief space. It contains the following information: (1) positions of best optima; (2) successful move lengths that was used to find these optima; and (3) promising area to search. • All individuals can contribute their knowledge to the belief space, but only the most successful one is used as exemplars for others to follow. Here are the details: − At the beginning of a generation, a promising region was created from positions of all best optima. Individuals are attracted to this region. E. Corchado et al. (Eds.): IDEAL 2006, LNCS 4224, pp. 586 – 594, 2006. © Springer-Verlag Berlin Heidelberg 2006
Hybridizing Cultural Algorithms and Local Search
587
− Each individual then uses rank selection to adopt with modification a successful move length from the belief space. − During its lifetime, an individual can change its move length by selecting with modification a new one from the belief space. • At each move, an individual can proceed along either a random direction or a direction heading to a selected optimum depending on a certain probability. To guarantee search diversity, the more the trials, the larger the probability is. Some details described in the framework above have already been mentioned in existing literature. The idea of iteratively taking the local search and using mutation to escape from current basin of attraction has been mentioned by Martin et al. [3], and then was named Iterated Local Search (ILS) by Lourenco et al. [4]. The idea of maintaining exemplars from current solutions and using them to influence the nextgeneration population was also described by Reynolds [5] in his Cultural Algorithms (CA). Our new algorithm is a hybrid version of CA and ILS. We named it “CA-ILS”. Several new issues are proposed in CA-ILS. Firstly, different from current CA versions for continuous unconstrained optimization [6, 7], which concentrate on evolving solutions, we tried to evolve behaviors that individuals can use to find better solutions. These include lengths, directions and regions for the move. Because there are evidences that in reality phenot
Data Loading...
