RE: Performance issues with OWL Reasoners => subclass vs instance-of

> I disagree and my point is that the universe you speak of is framed by a
> specific reasoning algorithm.  

[VK] I believe this is not true. It's been a while since I took courses in
Theoretical CS, but complexity classes are not based on a given technology. In
general these complexity classes are determined by either reducing a problem to
other problems that are known to fall in a given complexity class, typically
using set theoretic and graph theoretic approaches. A list of complexity classes
for various DL variants are given below.

http://www.cs.man.ac.uk/~ezolin/logic/complexity.html

So, as Ian pointed out, the way to attain efficiency would be to either
compromise on expressivity, thus bringing the problem into a lower complexity
class and more efficient or to optimize it for typical scenarios.

> But your point is taken (below) that
> experimentation and results are what is needed.  

Yes, experimentation is needed and that would help you identify scenarios in
which rule based approaches are more efficient and scalable than tableaux
reasoning or vice/versa.

> The reality is that the
> world of production systems and DL/FOL reasoning are somewhat isolated
> from each other and both can benefit greatly from the other.

[VK] That would be an interesting thing to investigate. I would believe that the
RuleML conference in conjunction with ISWC 2006 would address OWL/SWRL/RuleML
integration.

> I plan to.  I simply don't think the assumption that Tableau Calculus
> represents the known limitations of DL reasoning is a very useful one.

[VK] The limitations of DL reasoning are determined by complexity classes of
various "problems" such as concept satisfiability, classification, etc and there
have been multiple algorithms such as structural and tableaux algorithms to
address these problems.

---Vipul

Received on Tuesday, 19 September 2006 13:59:17 UTC