Posted by Anthony Bucci on April 28, 2007 at 11:08:23:
Hey all,
Sevan and I are slated to give the advanced coevolution tutorial at GECCO
this summer. I posted some materials from that here:
http://www.cs.brandeis.edu/~abucci/tut_gecco2007/
There are some PDF files, one of which includes an annotated bibliography.
I've also included the PowerPoint file if you want to lift slides.
As with the tutorial last year, Sevan and I tried to organize the mass of
material around some simple themes. Here we have 'interaction' and
'elaboration' as two primary themes in recent coevolution research. We
made what I think are useful connections between Sevan's notion of
solution concept/monotonic improvement on the one hand, and Ken Stanley's
notion of complexification on the other. Abstractly, both Sevan's and
Ken's ideas are about preserving the capabilities of solutions via an
elaborative process (where a mixture [as in Nash equilibria] or a set of
strategies [as in the non-dominated front] or a neural network topology
[as in NEAT] can all be viewed as attempts to accumulate already-found
capabilities).
Perhaps this conceptualization will allow a bridge between evaluation and
representation. I've always thought that what we've found about how we
evaluate things will eventually spill over into how we represent things.
Shiva's work on learnable test sets sketched how that connection might
look, but I think we can start to fill in the details now.
Anyway, comments welcome. Enjoy,
Anthony