Optimization methods usually have several user-defined parameters that govern the behavior and efficacy of the optimization method. These are called the optimizer's behavioral or control parameters. Finding a good choice of these behavioral parameters has previously been done manually by hand-tuning, and sometimes even by using coarse mathematical analysis. It has also become a common belief among researchers that behavioral parameters can be adapted during optimization to improve overall optimization performance; however, this has been demonstrated to be mostly unlikely. Tuning behavioral parameters can be considered an optimization problem and hence can be solved by an overlaid optimization method. This is known here as meta-optimization, but is also known in the chapter as meta-evolution, super-optimization, parameter calibration, and so on. The success of SwarmOps when doing meta-optimization relies mainly on the following three factors:

1. SwarmOps features an optimization method that is particularly suitable as the overlaid meta-optimizer because it quickly discovers well-performing behavioral parameters (this is the LUS method described in this chapter).
2. SwarmOps employs a simple technique for reducing computational time called pre-emptive fitness evaluation.
3. SwarmOps uses the same function-interface for both optimization problems and optimization methods. Several scientific publications use SwarmOps for meta-optimization and have more elaborate descriptions than those given here, as well as having literature surveys and experimental results. The concept of meta-optimization can be illustrated schematically as follows:

In the preceding diagram, the optimizer whose behavioral parameters are to be tuned is taken to the DE method, which we will look at later on in this chapter. The SwarmOps framework allows for parameters to be tuned regarding multiple optimization problems, which is sometimes necessary to make the performance of the behavioral parameters respond better to more general problems.

In the preceding example, the DE parameters are tuned for two specific problems.