Swarms, evolution and all that jazz

Collective Intelligence Systems draw their inspiration from biology where a number of simple units often achieve a performance that is greater than the sum of their parts because of the way they are organised together. The way that birds flock and perform efficient searches without any hierarchical organisation structure has inspired an algorithm called Particle Swarm Optimisation (PSO). In this talk it will be suggested that traditional PSO is still too biological and that with the addition of a non-biological feature and evolution it can be extended to perform efficient searches in highly complex problem domains. For example, in a problem domain with 10^27 discrete optima, it has a 40% chance of finding the global optimum with only 1.5*10^7 evaluations. Further more it will find a series of very good optima, not just one. This obviously has importance when the time taken to evaluate a possible solution to a problem is long, but the solution is needed fast. Many real world industrial problems have such constraints. The seminar will start with the basics of PSO before moving onto the new algorithm and showing the results on two complex problem domains. It will conclude with a suggestion (and hopefully preliminary results) as to how this remarkable performance can be made even better