ITK: ParticleSwarmInitializationBiasedOptimizer Class Reference

Implementation of a biased/regularized Particle Swarm Optimization (PSO) algorithm. More...

#include <itkInitializationBiasedParticleSwarmOptimizer.h>

Detailed Description

Implementation of a biased/regularized Particle Swarm Optimization (PSO) algorithm.

This PSO algorithm was originally described in: M. P. Wachowiak, R. Smolikova, Y. Zheng, J. M. Zurada, A. S. Elmaghraby, "An approach to multimodal biomedical image registration utilizing particle swarm optimization", IEEE Trans. Evol. Comput., vol. 8(3): 289-301, 2004.

The algorithm uses a stochastic optimization approach. Optimization is performed by maintaining a swarm (flock) of particles that traverse the parameter space, searching for the optimal function value. Associated with each particle are its location and speed, in parameter space. A particle's next location is determined by its current location, its current speed, the location of the best function value it previously encountered, the location of the best function value the particles in its neighborhood previously encountered and the initial position the user specified.

The assumption is that the user's initial parameter settings are close to the minimum, which is often the case for registration. The initial parameter values are incorporated into the PSO's update rules, biasing the search in their direction. The swarms update equations are thus:

$v_i(t+1) = wv_i(t) + c_1u_1(p_i-x_i(t)) + c_2u_2(p_g-x_i(t)) + c_3u_3(x_{init} - x_i(t))$ $x_i(t+1) = clampToBounds(x_i(t) + v_i(t+1))$

where $u_i$ are $~U(0,1)$ and $w,c_1,c_2, c_3$ are user selected weights, and c_3 is linearly decreased per iteration so that it is in $c_3=initial, 0$ .

Swarm initialization is performed within the user supplied parameter bounds using a uniform distribution or a normal distribution centered on the initial parameter values supplied by the user, $x_{init}$ . The search terminates when the maximal number of iterations has been reached or when the change in the best value in the past $g$ generations is below a threshold and the swarm has collapsed (i.e. particles are close to each other in parameter space).

Note:: This implementation only performs minimization.

The documentation for this class was generated from the following file:

itkInitializationBiasedParticleSwarmOptimizer.h