Re: GA ver. Parallel tempering

Dov wrote:
I have seen quite a few postings and studies comparing GA and SA
(simulated annealing).

That is a bit unfair comparison. GA uses a population, SA does not. A
more appropriate
comparison would be GA & parallel tempering (or replica exchange).

I'm not sure that one can say that it is unfair.

If you consider a cooling schedule using steps (where you sample several
points between a temperature decrease), SA is not so different than
populational metaheuristics. It just use a different probability density
function to generate the next iteration sample.

On the other side, if you use a GA where parameters changes during the
optimization, you are also close to a simulated annealing.

The right question to ask, from my point of vue, is "how do I build the
probability density function used to sample the objective function in the
next iteration ?".

Of course, this does not apply to a dynamic optimization problem

--
NoJhan
.