Re: learning with constraints
- From: "JGCASEY" <jgkjcasey@xxxxxxxxxxxx>
- Date: 27 Nov 2005 11:14:39 -0800
> JC:
> I suspect we can learn just by being
> exposed to an input with constraint as an
> automatic result of the way we are wired.
>
>
> GS:
> I'm sorry, this sounds like gibberish.
> Is that deliberate?
>
>
> JC: What I was thinking of was the acquisition
> of knowledge. For example a favorite beginner
> text processing program is to count the
> frequency of each character in a text string.
> Then to count how frequently one character
> is followed by another.
>
>
> Thus we can say the program has *learnt* how
> frequently each character occurs and has
> *learnt* how often each character is followed
> by each other character.
>
>
> GS:
> You can call a sow's ear a "silk purse"
> and putting it in quotes doesn't make it
> any less stupid.
It would mean it was still a purse, maybe
a very good purse, just not one made of silk.
At the start of the program, before it is fed
the text, its "behavior" would be random in
terms of predicting the next character. After
enough text its behavior would be less random
and most people would say it has learnt. How
would you describe the change in behavior?
> JC: This learning is the result of how the system
> (computer) is wired (program) and being given
> (exposed to) an input with constraint.
>
>
> GS:
> Why worry about wiring? After all, if I put a
> rain gauge out in my yard it "learns" how much
> rain has fallen. No? Yeah, that's the ticket!
> It "knows" how much rain has fallen.
>
>
> What is wrong with you that you cling so
> tenaciously to such ridiculous notions?
Because they work when applied to real systems.
Someone with imagination would think, how can
I make use of the "knowledge" in the rain gauge.
You on the other hand might think, how ridiculous,
and walk of in disgust.
An action requires something to act upon. In
this case it is internal knowledge. An internal
change that reflects some outside event which
in turn effects the behavior of the system.
Now imagine your rain gauge is on one end of
a lever. When the rain gauge is empty it will
turn on the irrigation system. At some point
the water falling into the rain gauge will
reach a weight that will drop the lever and
turn off the irrigation.
Such systems exist and the advantage is they
will adjust the rate of watering according to
the amount of evaporation (from the rain gauge
and thus from the plants) and the amount of
watering from rain or irrigation.
The "knowledge" about how much watering from
rain or irrigation and how much evaporation
has taken place due to wind/heat is contained
in the rain gauge.
To an objective observer the system "knows"
when to water.
> JC:
> An input without constraint would make every
> character and sequence of characters equally
> likely and therefore what would there be to
> learn?
>
>
> GS:
> There's no "constraint" on the input in
> the original.
So what does "constraint" mean to you? Objects
are defined and recognized by their constraints.
> JC:
> There is no reward system involved, it is just
> the consequence of the wiring (program) and
> the type of input (non random).
>
>
> GS:
> You must have left something out of the original
> description. The input could be random and the
> program could still count the frequency of each
> letter and the conditional probability of each
> letter following every other letter. But who
> cares? The whole thing is ridiculous.
Yes I did leave something out of the original in
that it was the "knowledge" acquisition part of
the system only. Yes indeed it would have counted
the frequencies even if they were random. In a
normal conversation much is left out with the
assumption the person reading it will be able to
get what point is being made without having to
spell it out.
If you were looking at a detuned tv screen with
just random noise you would be in exactly the
same situation in describing what was on tv.
Knowledge acquisition is often used as a measure
of intelligence. Who is President, what time of
day is it, what is the color of a leaf, what is
Einstein's famous equation ....
When people acquire knowledge they are said to
have learnt these things even though in terms
of being able to use this knowledge creatively
they may be completely useless.
--
JC
.
- References:
- learning with constraints
- From: JGCASEY
- Re: learning with constraints
- From: Glen M. Sizemore
- learning with constraints
- Prev by Date: Re: Cool visual illusion
- Next by Date: Re: Cool visual illusion
- Previous by thread: Re: learning with constraints
- Next by thread: Re: learning with constraints
- Index(es):
Relevant Pages
|
