Good day.
I have been playing around with Evolutionary Programming (Evolutionary Algorithm variation) lately, specifically self-adaption where the strategy parameters evolves with the solution.
With the additive technique, the theory gives (Fogel, et al, 91):
Here sigma is the strategy parameter with eta the learning rate and N() some normal distribution of random numbers to adapt the strategy parameter with the solution. It makes sense as this is clearly a form mutation. You use the existing strategy parameter and changes it slightly.
With multiplicative technique, the theory gives (Jiang and Wang, 05):
With t the current iteration, n_t the maximum allowed and the lambdas all static constants. The previous value for the strategy parameter is not even used, but only the initial value, sigma(0).
This does not make sense to me at all. There is no random component and the strategy parameter is deterministic. Can someone please shed some light here? How does this adapt with the solution? This feels more like a "dynamic" approach.
