Swarm Intelligence

Appendix A

Statistics for Swarmers

S

tochastic algorithms, including evolutionary methods, some gradi-

ent descent algorithms, simulated annealing, and particle swarms—

most of the approaches discussed in this book—can perform differently

each time they are run. If a landscape has a number of hills, the algo-

rithm may start in a better or a worse region and may proceed from there

to a better or worse point in the problem space. A researcher is often

interested in comparing the effects of various parameter levels, or differ-

ent algorithms, or problem types, or some other causal factor, called an

independent variable, on some effect or dependent variable, ...