Re: GA when dimension of problem is huge, GA + classification

Hi Artem:

Regarding your 1st question, many approaches are possible to try to
overcome this problem. Once, I had a huge search universe, on a problem
related to the classification of digital images (specifically a colour
segmentation problem, which is a little bit different) with different
layers of information coming from satellite images. In order to
overcome the huge search space I have tried Neoteny:
http://alfa.ist.utl.pt/~cvrm/staff/vramos/Vramos-EvoIASP01.pdf, which
can be simply applied: collect primitive DNA (chromosome material) in
the first generations, and re-inject it at the long run. This simply
process can ensure diversity over the GA population and lead to faster
convergence. Other approaches, include, for instance, varying
populations for each GA generation. Last but not (certainly) least is
to have a look on the Back & Fogel Evol. Comp. handbook. There, you
will find detailed approaches.

Regarding your 2nd question (classification), just imagine that you
have an classification problem (trying to map from feature data from
many items to a class decision). Here, there are also many approaches,
many of it are classical and quite robust. However, you can look at the
problem as an optimization problem. If that's the case, you could apply
AG's. For one example have a look on:
http://alfa.ist.utl.pt/~cvrm/staff/vramos/ref_13.html.
Also you can hybridize EC with know standard techinques as, for
instance, K-means.

Best, vitorino

.