Ten theses regarding the design of controlled evolutionary strategies

Resume

Two issues highlight the differences between this kind of strategy design and other evolutionary approaches:

• 1) The evolutionary processes of nature have model character only. It is not aimed at exactly simulating nature's mechanisms with the computer (cf. genetic algorithms). Any attempt at exactly simulating natural processes with the computer would require too much effort in terms of calculating and thus not be in line with our target.

  It is rather aimed at transferring nature's evolutionary mechanisms to a computer program so that they produce clear advantages in terms of efficiency. I. e., selection is done under very pragmatic aspects, and if nature does not provide any patterns for the specific problem, the designer is called upon to create new evolutionary rules.

• 2) The design of controlled evolutionary strategies is not limited by the level of complexity which still alows mathematical analysis of the strategy's convergence quality.

  The module approach rather allows the design of strategies — based on empirical tests — wich employ 5 and more adaption processes at one time and which are surely too complex to be analysed with mathematical tools.

There is reason to believe that particularly the complex optimization problems occuring in practice can be approached with controlled evolutionary strategies.