On 1/29/08, Terry Hancock <[EMAIL PROTECTED]> wrote:

> PhD candidates do exist. ;-)

Indeed, and if we can frame this FPGA synthesis problem in terms of
either cognitive science or abductive inference, maybe I can get my
advisor to let me do this for my dissertation.  :)

Typically, simulated annealing, genetic algorithms, or some other kind
of randomized hill-climbing method is applied to this problem.
However, if we can rephrase it in terms of hypothesis
generation/matching and aggregation, maybe we can get somewhere.  I'm
not sure it's a good fit.  Part of the problem here is that it IS a
'solved' problem, in the sense that halfway reasonable solutions have
been produced before.  Doing the same thing again may be worthy of a
Master's thesis.  But for a Ph.D., you have to invent something new.
Provide a new insight into the problem that helps us deal with the
problem in a better way or teaches us something we didn't know before.
 This is why, for me to work on it as my main line of research, it
would have to be 'better' and not just a reimplementation of
well-understood techniques.  Even just doing a better job with old
techniques isn't enough.

All AI is about search, just different spaces and different
techniques.  VLSI synthesis is about search for a good placement and
routing.  With abduction, we're typically searching for a best
explanation for some observation.  It's like deduction in reverse.
Deduction tells you that if something is a dog, then it barks.
Abduction tells you that if you hear barking, that's evidence for the
presence of a dog.  Abduction subsumes induction.  How can we describe
a logic gate placement as the best explanation for a given set of
unplaced gates?  :)

Beats the heck out of me...  :)

> For our purposes, of course, "better" is nice but not obligatory. I
> don't imagine anyone expects a free synthesis tool (right term?) to beat
> a proprietary one on its first outing. We'd just expect it to "work", right?

Very true.

-- 
Timothy Normand Miller
http://www.cse.ohio-state.edu/~millerti
Open Graphics Project
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to