From: James Rogers (jamesr@best.com)
Date: Thu Nov 06 2003 - 23:12:42 MST
On 11/1/03 2:00 PM, "Perry E.Metzger" <perry@piermont.com> wrote:
> "Metaqualia" <metaqualia@mynichi.com> writes:
>> I was thinking of a universal pattern recognition machine, something that
>> will work to recognize whatever, no matter the stimulus. it will recognize
>> automatically configurations with low entropy and high organization in the
>> sensory input. The same machine could learn to do audio or video analysis or
>> solve problems, whatever.
>
> In the general case, what you're talking about would violate Rice's
> theorem. What you're looking for would distinguish non-trivial
> properties of the R.E. languages, so it is a priori impossible.
This depends on the assumptions to a certain extent. If we are talking
about general non-axiomatic induction, a function solution (not THE
solution, just a GOOD approximation) can always be found for these types of
cases, and with a known error bound. The only classes of function that are
not discernable in this case are also not within the scope of the original
query. Actual correct solutions are too expensive for non-trivial cases.
For FSMs trying to discern universal patterns, the old 80/20 rule is
optimal, so hopefully those kinds of solutions are good enough. Not too
many of us have the luxury of reversing functions with a genuine UTM.
-James Rogers
jamesr@best.com
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:43 MDT