ISO group for expert systems

AFAIK there isn't any expert systems groups. I have been using expert systems in a lot of the work that we do in code generation and fuzzy logic systems. Expert systems are very application specific. There are a great diversity of goals and approaches and most of what one application needs is not applicable to another.

Some of the useful terms are artificial intelligence, fuzzy logic, linguistic variables, and constraints. All of these have algorithms useful in building expert systems.

Expert systems implementation needs as much attention as the application theory. Implementation can have a very big impact on system performance.

Regards,

-- Walter Banks Byte Craft Limited

Sadly, that appears to be the case. THough I am pursuing some email contacts which may pan out...

Of course! Fuzzy logic vs. backward chaining inference, etc.

As I said, I'm just looking for a simple production system. Forward chaining, exclusively. Low data rate but *memory* considerations outweigh speed (the reverse of what is usually the case).

Yes. And, with a large KB, this can be significant. Especially when "hobbled" with my memory constraints.

Currently, my thinking is to isolate the conditions within each rule (as rules are entered -- incremental compiler). Uniquely identify conditions and rules. Link conditions to the rule(s) that reference them.

Similarly, identify and link *facts* to the condition(s) that reference *them*.

Thereafter, asserting a fact should give me a constrained list of conditions to (re)evaluate. THese, in turn, give a constrained set of rules to fire. etc.

I think -- with careful consideration of how a condition is "defined" (e.g., a compound condition might be better served if represented as a set of several *simple* conditions) and locality of reference among conditions, rules, references, etc. -- that I can compensate for the constraints placed on the memory available.

So, my goal is to find others with *practical* experience "under the hood" to determine where the bottlenecks are likely to manifest.

The KB is a different issue :>

Not aware of any ISO groups. You could look around in the comp.ai branch but AFAIHS their traffic is sporadic these days.

You might do better to poke around in language forums. A lot of rule based systems are implemented in Lisp, Prolog, Poplog, OPS5, etc. and you may find people in those forums who have practical experience implementing them. It may be beneficial even if you intend to work in a different language.

George

"A regular commedian!" ;-)

Seems to be the norm for the more technical groups :<

Not really looking for help implementing the expert system, per se. Rather, the inference engine (on which to build the system).

I found a CLIPS "google group" but don't think I want to bother with All Things Google... :<

Well, your original post said "expert systems", not inference engines.

Anyway, is this an academic quest or do you have a reason not to start with one of the existing open source engines? Where to start depends on whether you need forward or backward chaining inference (or both).

I'm reasonably sure there are *NO* groups on inference engine design! :>

From my reply to another post:

"As I said, I'm just looking for a simple production system. Forward chaining, exclusively. Low data rate but *memory* considerations outweigh speed (the reverse of what is usually the case)."

Yes.

From what I saw, there, it was mostly concerning the designs of the expert systems themselves -- and how to exploit features of CLIPS to that end, etc.

I've pulled the documentation for CLIPS and am teasing out the details of the implementation -- coupled with other references regarding Rete, etc. -- to determine how much I can "bend" an implementation and it's likely consequences in terms of performance.