Locust: A parallel multi-objective particle swarm optimizer (PSO)
Locust is an innovative multi-objective particle swarm optimizer, and the perfect counterpart to Ferret. Locust is a much smaller code than Ferret, which is based on the collective intelligence of a swarm of interacting particles. It skips much of Ferret's high-powered linkage-learning analysis, but still performs extremely well even on very difficult problems.
How is this possible? Think of the behavior of a school of fish, a flock of birds, or a swarm of insects. Intelligent, complex behavior emerges as a remarkable property of the system as a whole, from the interaction of simple component parts. Locust takes advantage of this property and uses it as the basis of a multi-objective optimizer.
The second generation Locust code is almost as powerful as Ferret, and we consider it to be one of the “front-line” optimizers of the Qubist package. Locust has relatively few options or settings, but it is a full-featured, parallel code with the same visualization interface as Ferret. Locust should be seen as complementary to Ferret, as a smaller, “snappier” code with excellent search and optimization properties but less internal overhead.
Features:
- Multi-objective analysis: Locust is a powerful multi-objective optimizer, which efficiently maps trade-off surfaces. Regions of parameter space can also be mapped for single-objective problems when a “fuzzy tolerance” is defined on the objective function.
- Exact particle dynamics: Locust is unique as a PSO because it is based on an exact analytical solution to the equations governing each particle step, which improves the stability of the algorithm.
- Sub-swarms: Multi-objective optimization and parameter space mapping use a novel algorithm based on a dynamic local-best swarm topology. An evolving group of “sub-swarms” each map part of the trade-off surface or optimal set. Sub-swarms move, merge, and divide automatically to optimize their coverage of the optimal set.
- Advanced analysis, visualization, and parallel computing: exactly the same powerful features as Ferret.
