Re: Calculemus

From: Christian Szegedy (szegedy@t-online.de)
Date: Wed Nov 10 2004 - 16:40:30 MST


> This is interesting indeed, but I don't know of any work that makes this
> practically useful, do you? If so I'd love to hear about it.

I don't see its immediate relevance for AI. Note that the PCP is about
checking of the correctness of proofs, not of conjectures.

The constants in the PCP theorem are not astonomical and the degrees
of the polynomials are small, so it could be practically feasible.

The significance of the PCP theorem is mainly theoretical today: it
is the base of almost all inapproximability results.

> Yeah, but control theory totally falls apart when you start dealing with
> situations of significant complexity.

This sounds like an ad-hoc statement. Of course, as someone working on
CAD systems for several years, I experience the gap between mathematically
provable and practically efficient every day. Still, I also see that
mathematically justified methods and mathematical insights lead to
much better performance after additional tricking around than
pure heuristics.

> Is there more online information about Bless somewhere?

BLess is in the implementation phase. The ideas are very
stable, but there is no documenation yet. As soon as I release
BLess 0.1, there will be a documenation.
The only currently available resource is a short discussion
on comp.lang.functional:

http://www.talkaboutprogramming.com/group/comp.lang.functional/messages/49357.html



This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:50 MDT