First, I apologize for sending this response by e-mail. I hit the wrong
button. I generally try to avoid e-mail without invitation.
*************************************
"Harry Erwin" <herwin at gmu.edu> wrote:
>>> You're missing my point. Symbols are signs. They belong to a countable
>>> set. Wind-tunnel models can vary continuously (or discontinuously). That
>>> matters--there are some applications (for example in hydraulic analysis)
>>> where symbolic modeling encounters an intractable problem, but analog
>>> modeling works fine.
>> Keeping in mind that I am still very new to all this, I am not sure I
>> understand the distinction you are trying to make between "symbolic
> > modeling" and "analog modeling". Isn't the comparison of "symbolic" to
>> "analog" something along the line of a comparison of metrics to
>> precision?
> Perhaps I would be clearer if I used the term 'formal model' rather than
> 'symbolic model'.
Not really; but please keep in mind my limited knowledge of the subject.
"Formal model" to me seems pretty much synonymous with "symbolic model".
The distinction between this and "analog modeling" still confuses me. I
mean, wouldn't an "analog model" simply be a "formal model" in which
symbols, functions, and so forth are allowed to take on real values?
There are formal models for analog recursive neural nets (ARNN). Hava T.
Siegelmann offers a really good overview of this area of research.
>>> Why do I care? Disambiguating an acoustic scene
>>> based on multisensor data is very difficult because of all the ghosts
>>> that have to be eliminated _sequentially_. Bats do it in real time. How?
>> This also may sound simplistic, but why sequentially? Can't the bat
>> focus on specific acoustic patterns? They do possess the potential
>> for handling pattern recognition and pattern definition. And, they
>> also possess the potential for focused attention.
> The matching problem is NP-complete even assuming no noise.
You will get no argument from me here.
