Stefan Janson

A hierarchical particle swarm optimizer and its adaptive variant

A hierarchical version of the particle swarm optimization (PSO) metaheuristic is introduced in this paper. In the new method called H-PSO, the particles are arranged in a dynamic hierarchy that is used to define a neighborhood structure. Depending on the quality of their so-far best-found solution, the particles move up or down the hierarchy. This gives good particles that move up in the hierarchy a larger influence on the swarm. We introduce a variant of H-PSO, in which the shape of the hierarchy is dynamically adapted during the execution of the algorithm. Another variant is to assign different behavior to the individual particles with respect to their level in the hierarchy. H-PSO and its variants are tested on a commonly used set of optimization functions and are compared to PSO using different standard neighborhood schemes.

Honeybee swarms : how do scouts guide a swarm of uninformed bees?

The organized movement of a swarm of honeybees towards its new home is a perplexing phenomenon because only a small number of scout bees, approximately 5%, know the direction in which the swarm has to move. Nevertheless, in the majority of cases a swarm, comprising about 10 000 mainly uninformed bees, reaches the new home. How do the scouts transfer directional information en route to the uninformed bees? We investigated a hypothesis proposed in the 1950s that suggests that scout bees fly rapidly through the airborne swarm, pointing towards the new home. We developed a model that simulates the movement of swarms and scouts and showed that when scouts fly through the swarm at a speed slightly higher than the speed of the other (uninformed) bees, they are indeed able to direct the swarm towards its new home. Hence, our model strongly supports the proposed hypothesis and shows that a collection of uninformed bees can be successfully guided by the purposeful movements of a small number of informed scouts.

A hierarchical particle swarm optimizer for noisy and dynamic environments

New Particle Swarm Optimization (PSO) methods for dynamic and noisy function optimization are studied in this paper. The new methods are based on the hierarchical PSO (H-PSO) and a new type of H-PSO algorithm, called Partitioned Hierarchical PSO (PH-PSO). PH-PSO maintains a hierarchy of particles that is partitioned into several sub-swarms for a limited number of generations after a change of the environment occurred. Different methods for determining the best time when to rejoin the sub-swarms and how to handle the topmost sub-swarm are discussed. A standard method for metaheuristics to cope with noise is to use function re-evaluations. To reduce the number of necessary re-evaluations a new method is proposed here which uses the hierarchy to find a subset of particles for which re-evaluations are particularly important. In addition, a new method to detect changes of the optimization function in the presence of noise is presented. It differs from conventional detection methods because it does not require additional function evaluations. Instead it relies on observations of changes that occur within the swarm hierarchy. The new algorithms are compared experimentally on different dynamic and noisy benchmark functions with a variant of standard PSO and H-PSO that are both provided with a change detection and response method.

A hierarchical particle swarm optimizer

A hierarchical version of the particle swarm optimization method called H-PSO is introduced. In H-PSO the particles are arranged in a dynamic hierarchy that is used to define a neighborhood structure. Depending on the quality of their so far best found solution the particles move up or down the hierarchy so that good particles have a higher influence on the swarm. Moreover, the hierarchy is used to define different search properties for the particles. Several variants of H-PSO are compared experimentally with variants of the standard PSO.

A new multi-objective particle swarm optimization algorithm using clustering applied to automated docking

In this paper we introduce the new hybrid Particle Swarm Optimization algorithm for multi-objective optimization ClustMPSO. We combined the PSO algorithm with clustering techniques to divide all particles into several subswarms. Strategies for updating the personal best position of a particle, for selection of the neighbourhood best and for swarm dominance are proposed. The algorithm is analyzed on both artificial optimization functions and on an important real world problem from biochemistry. The molecule docking problem is to predict the three dimensional structure and the affinity of a binding of a target receptor and a ligand. ClustMPSO clearly outperforms a well-known Lamarckian Genetic Algorithm for the problem.

On Trajectories of Particles in PSO

The moving behaviour of the particles in particle swarm optimization (PSO) algorithms is studied in this paper. It is shown that particles in standard PSO have a clear bias in their movement direction that depends on the direction of the coordinate axes. This has the effect that the optimization behaviour of standard PSO is not invariant to rotations of the optimization function. A second problem of standard PSO is that non-oscillatory trajectories can quickly cause a particle to stagnate. A sidestep mechanism is proposed to improve the movement of the particles. A particle performs a sidestep with respect to a certain dimension when stagnation of movement along this dimension is observed. It is shown for simple test functions that the movement behaviour of sidestep PSO can prevent the unwanted bias and makes PSO less dependent on rotations of the optimization function. It is also shown for standard benchmark functions that sidestep PSO outperforms standard PSO

Hardware-oriented ant colony optimization

A new kind of ant colony optimization (ACO) algorithm is proposed that is suitable for an implementation in hardware. The new algorithm - called Counter-based ACO - allows to systolically pipe artificial ants through a grid of processing cells. Various features of this algorithm have been designed so that it can be mapped easily to field-programmable gate arrays (FPGAs). Examples are a new encoding of pheromone information and a new method to define the decision sequence of ants. Experimental results that are based on simulations for the traveling salesperson problem and the quadratic assignment problem are presented to evaluate the proposed techniques.

Hierarchical cellular genetic algorithm

Cellular Genetic Algorithms (cGA) are spatially distributed Genetic Algorithms that, because of their high level of diversity, are superior to regular GAs on several optimization functions. Also, since these distributed algorithms only require communication between few closely arranged individuals, they are very suitable for a parallel implementation. We propose a new kind of cGA, called hierarchical cGA (H-cGA), where the population structure is augmented with a hierarchy according to the current fitness of the individuals. Better individuals are moved towards the center of the grid, so that high quality solutions are exploited quickly, while at the same time new solutions are provided by individuals at the outside that keep exploring the search space. This algorithmic variant is expected to increase the convergence speed of the cGA algorithm and maintain the diversity given by the distributed layout. We examine the effect of the introduced hierarchy by observing the variable takeover rates at different hierarchy levels and we compare the H-cGA to the cGA algorithm on a set of benchmark problems and show that the new approach performs promising.

On Enforced Convergence of ACO and its Implementation on the Reconfigurable Mesh Architecture Using Size Reduction Tasks

In this paper we show that size reduction tasks can be used for executing iterative randomized metaheuristics on runtime reconfigurable architectures so that an improved throughput and better solution qualities are obtained compared to conventional architectures that do not allow runtime reconfiguration. In particular, the problem of executing ant colony optimization (ACO) algorithms on a dynamically reconfigurable mesh architecture is studied. It is shown how ACO can be implemented such that the convergence behavior of the algorithm can be used to dynamically reduce the size of the submesh that is needed for execution. Furthermore we propose a method to enforce the convergence of ACO leading to a faster reduction process. This increases the throughput of ACO algorithms on runtime reconfigurable meshes. The increased throughput is used for repeated runs of ACO algorithms on a given set of problem instances which significantly improves the obtained solution quality.

A decentralization approach for swarm intelligence algorithms in networks applied to multi swarm PSO

Purpose – The purpose of this paper is to present an approach for the decentralization of swarm intelligence algorithms that run on computing systems with autonomous components that are connected by a network. The approach is applied to a particle swarm optimization (PSO) algorithm with multiple sub‐swarms. PSO is a nature inspired metaheuristic where a swarm of particles searches for an optimum of a function. A multiple sub‐swarms PSO can be used for example in applications where more than one optimum has to be found.Design/methodology/approach – In the studied scenario the particles of the PSO algorithm correspond to data packets that are sent through the network of the computing system. Each data packet contains among other information the position of the corresponding particle in the search space and its sub‐swarm number. In the proposed decentralized PSO algorithm the application specific tasks, i.e. the function evaluations, are done by the autonomous components of the system. The more general tasks...