Main Page | Report this Page
Computers Forum Index  »  Computer Artificial Intelligence - Philosophy  »  how do neurons code information...
Page 1 of 1    

how do neurons code information...

Author Message
casey...
Posted: Thu Aug 06, 2009 9:36 pm
Guest
Curt wrote:
Quote:
In a pulse langauge, zero information is encoded
in the symbol selection. That's because there's
only one symbol in the language - the pulse.

There is only one symbol in a computer as well but
groups of bits in space or time can code letters
and words. There is also place coding so a pulse
on one line can mean one thing and a pulse on
another line can mean something else.

Quote:
All the information is instead, encoded in _when_
it's sent - because that's the only way to encode
information in a such a langauge.

It is not clear to me what is being coded in "when"
a pulse is sent. How does it know when it was sent?
A position in time or space requires an origin.

In your pulse sorter there was a duration of time,
not a when it arrived, only how much time had passed
since the last pulse.

In your bog standard integrate and fire model of a
neuron it will fire whenever it gets enough positive
juice which is being added and subtracted by many
excitatory or inhibitory inputs. When those pulses
arrive is not computed only that more positive than
negative pulses _have_ arrived _in time_ to push
the neuron into firing.

In the ordinary use of the word "when" if I ask
when did you send the letter you might say last
Thursday at six thirty-five pm.

What "information" is coded in the "when" you
sent the letter?

A clear example with some neural model might clear
up the notion you are presenting. I have played with
many plausible neuron like models and am at a loss
as to exactly what you might mean here by "when"
a pulse is sent or even what information you see
being coded in that sent time?


JC
 
casey...
Posted: Fri Aug 07, 2009 5:32 pm
Guest
On Aug 7, 12:20 am, c... at (no spam) kcwc.com (Curt Welch) wrote:
Quote:
If you take some sequence of multiple inputs to a
neuron, and see that some sequence of 100 inputs,
finally causes the neuron to fire on input pulse
number 100, you can go back in time, and change
just the arrival time of some of those pulses, and
it will change when the neurons fires. It will
either cause the neuron to fire sooner, or cause
it not to fire on pulse 100 - not because you
changed the weights of any inputs, but just
because you adjusted when the pulses arrived.

Firing sooner or later is not encoding anything in the
example above except to say _when_ it happens.

If neuron A1 in a moth's ear is firing it tells the
moth that there is a BAT present NOW. There is only
one piece of time information and that is the event
is happening NOW. _What_ is happening now is encoded
in _which_ neuron is firing not _when_ it is firing.

It uses the place coding which you seem to neglect.

The wiki reference you gave that covers different coding
schemes *doesn't talk that way*. It doesn't say anywhere
that "information is coded in the _when_ a pulse occurs".

Quote:
The brightness of the light is not encoded in some
sequence of 1's and 0's. It's encoded in _when_ you
send the pulse.

That is silly talk. It is encoded in *how many pulses are
generated per second* but _when_ those pulses are generated
codes for nothing. It is a silly way to talk about frequency
coding, or other forms of temporal coding, (which I understand
perfectly well), for it implies a time stamp.

If you have a reference clock you could code things by when
they occur but that isn't how the neuron does it. Yes it
uses temporal coding but it doesn't use a particular "when"
(a *particular* point in time) to code anything.

Quote:
Because pulses don't have 2 values, the only way to
send information with them is as a temporal code.

The value is in the pulse being there or not.

The input of a neuron does have two values. It either
has a pulse present at an input or it doesn't have a
pulse present. Functionally it is no different to the
input of any binary digital unit.

Quote:
They don't carry any information in the spatial domain.

They DO carry information in the spatial domain by which
neurons are being used. They may carry other types of
information in a temporal code such as the intensity of
the stimulus.

Quote:
Do you want me to make up more examples of temporal
codes for you?

No point if you are pretending it explains why you say
it is in the _when_ a pulse arrives which in normal
usage means a time stamp or go on ignoring the place
code of a pulse or pretending that the absence of a
pulse at the input of a neuron codes for nothing. Or
pretend you think I don't understand temporal coding
and give a lot of examples to show "the idiot" what it
means when the issue was your sloppy use of language
to try and imply "thinking temporal" has some kind of
magic to it with your "when" talk.


JC
 
Curt Welch...
Posted: Sun Aug 09, 2009 5:15 am
Guest
casey <jgkjcasey at (no spam) yahoo.com.au> wrote:
Quote:
On Aug 7, 12:20=A0am, c... at (no spam) kcwc.com (Curt Welch) wrote:
If you take some sequence of multiple inputs to a
neuron, and see that some sequence of 100 inputs,
finally causes the neuron to fire on input pulse
number 100, you can go back in time, and change
just the arrival time of some of those pulses, and
it will change when the neurons fires. It will
either cause the neuron to fire sooner, or cause
it not to fire on pulse 100 - not because you
changed the weights of any inputs, but just
because you adjusted when the pulses arrived.

Firing sooner or later is not encoding anything in the
example above except to say _when_ it happens.

I've tried to explain this stuff to you in every way I can think to explain
it. I've made up endless examples, and made them as simple as possible,
but yet, you still don't seem to understand what temporal encoding of
information is all about.

Quote:
If neuron A1 in a moth's ear is firing it tells the
moth that there is a BAT present NOW. There is only
one piece of time information and that is the event
is happening NOW. _What_ is happening now is encoded
in _which_ neuron is firing not _when_ it is firing.

It uses the place coding which you seem to neglect.

I've been talking about the fact that pulse signals encoding information in
two very important ways for 5 years here now John. One way, is temporal
encoding, the other is spatial encoding. 5 years I've written long and
detailed messages about all this - giving endless examples to make sure
anyone that cared, could follow what I was talking about. Multiple times,
when Louis (aka Traveler) would claim that his networks were purely
temporal, I would point out to him that it was both spatial, and temporal.
This is very old, and very long debate I've been having here.

But yet, you seem unable to understand that all this talk about spatial
encoding, IS what you want to call "place coding". Do you understand
nothing? If I don't use the word "place coding" you think I'm ignoring the
fact that data is encoded in _which_ pulse was sent? How can this trivial
stuff fly so far over you head?

Quote:
The wiki reference you gave that covers different coding
schemes *doesn't talk that way*. It doesn't say anywhere
that "information is coded in the _when_ a pulse occurs".

How I talk is not important John. As long as I'm being consistent, you
should be able to figure it out what I'm talking about. I shouldn't have
to use the term "place code" before you recognize the _concept_ behind the
words. As I've said before, your strong suit has never been abstraction -
which is the ability to recognize the concepts behind the words. If I
don't use the exact same words that you read about somewhere else, you
don't seem to have the ability to realize I'm talking about the same effect
- the same abstract concept.

And this seems to be the real problem of you not understanding what "when
the pulse fires" mean or why it's significant. You can understand it if
someone calls it "frequency encoding", but you can't seem to grasp that
it's impossible to do frequency encoding without at the same time, using
_when_ the pulse is sent to encode information.

Quote:
The brightness of the light is not encoded in some
sequence of 1's and 0's. It's encoded in _when_ you
send the pulse.

That is silly talk. It is encoded in *how many pulses are
generated per second* but _when_ those pulses are generated
codes for nothing.

This stuff just seems to be beyond what you can understand. You probably
shouldn't waste you time reading my messages - they will only continue to
confuse you.

--
Curt Welch http://CurtWelch.Com/
curt at (no spam) kcwc.com http://NewsReader.Com/
 
Curt Welch...
Posted: Sun Aug 09, 2009 6:12 pm
Guest
Don Stockbauer <donstockbauer at (no spam) hotmail.com> wrote:
Quote:
On Aug 9, 2:18=A0am, casey <jgkjca... at (no spam) yahoo.com.au> wrote:
On Aug 8, 8:01=A0pm, c... at (no spam) kcwc.com (Curt Welch) wrote:



But yet, you seem unable to understand that all this
talk about spatial encoding, IS what you want to call
"place coding". =A0Do you understand nothing?

And what dumb arse response it that? What makes you write
that? When have I ever said a spatial code and place code
were not the same thing? You make these crappy statements
all the time pretending I believe or think things that I
have never made reference to. I was responding to this:

ALL the information is instead, encoded in _when_
it's sent - because that's the only way to encode
information in a such a langauge.

You wrote ALL, not some, is coded in "when it is sent" and
did not write that some is coded a spatial (place) code.

As I've said before, your strong suit has never been
abstraction - which is the ability to recognize the
concepts behind the words.

No. Concepts are learned from examples from which concepts
are *abstracted*. You play games, perhaps without realising,
with words. You are not using high level abstractions you
are playing word games.

If I don't use the exact same words that you read about
somewhere else, you don't seem to have the ability to
realize I'm talking about the same effect - the same
abstract concept.

And I would suggest you fool yourself this way by using
words meant for one level of explanation at another level.

... but you can't seem to grasp that it's impossible to
do frequency encoding without at the same time, using
_when_ the pulse is sent to encode information.

Rubbish. When something happens is equivalent to where
it happens *with respect to giving it an origin*.

A spatial or temporal pattern can exist at ANY time.

A burst of pulses of say 50Hz is independent of when they
occur. Of course they occur at some time but that isn't
relevant to the fact they code something in the frequency
of the pulses NOT in when these frequencies occur. If I
say it is a frequency of 50Hz I don't have to say it
occurred at 6:34 pm 10th May 2001 for it to be understood.

If _when_ something is used to code something it is
*different* to coding something as a frequency of pulses.
It is conflating the concepts to think frequency coding
needs a when to make sense.

This stuff just seems to be beyond what you can
understand. You probably shouldn't waste you time
reading my messages - they will only continue to
confuse you.

I think you confuse yourself. I wondered if you really
had something special to say but whenever you explain
yourself I see it as not something special but just an
eccentric way to say something or using a personal
meaning of a word which gives the impression that it is
something new or insightful.

There is nothing for example in the wiki reference you
gave that I found confusing. Nothing in any of the
books I have read on neurons, pulse coding, reinforcement
learning and so I that I have found confusing. Never
have I felt the need to chase up an explanation on any
of those topics. So what does that tell you?

I guess I have been a fool thinking you might have some
view worth understanding when in fact there was nothing
there to understand.

JC

Are yall going for some sort of record here?

Yes, but the problem is, we don't know what the old record is!

--
Curt Welch http://CurtWelch.Com/
curt at (no spam) kcwc.com http://NewsReader.Com/
 
casey...
Posted: Sun Aug 09, 2009 8:47 pm
Guest
On Aug 9, 11:22 am, c... at (no spam) kcwc.com (Curt Welch) wrote:
Quote:


When you say 50 Hz signal, into my mind, pops some real
world example of a 50 Hz signal. And these real world
examples always do happen at a precise point in time -
even though you didn't specify the point when you called
it a 50 Hz signal.

It is your thinking about the real continuos world you
believe is out there that confuses you when it comes to
how brains work. That continuos world out there is in
fact turned into temporal and spatial objects as a means
of dealing with it effectively. Yes we also deal with
the continuos and how it all fits together is reflected
in how we talk about these things. Language is a mirror
on how the human mind works not simply a result of
conditioning.

The precise point in time is not important if the information
is in the frequency rather than in a particular arrival time.

It is the same as a spatial pattern. It doesn't matter where
precisely in space this spatial pattern resides in order to
label the pattern. In fact the absolute spatial position is
separated out as useless information in recognizing the
pattern both in software I have written and in the brain
that has two separate pathways for the "what" and "where" of an
object. It is possible to know where an object is without
being able to identify the object and it is possible to
know what an object is without knowing where it is.

Quote:
... they all occur at precise points in space and time -
even though we choose to only talk about the frequency
attribute of that real thing.

Because the _information_ being _used_ is in the frequency of
the pulses not in their precise position in space or time.

If the information is coded in the frequency then the time
of arrival is not part of that information. You were saying
the information was in the _when_ the pulses arrived. I am
saying it is in the _frequency_ of arrival of the pulses.

You _could_ use time of arrival to code information and the
brain may even use that where useful but it would require a
reference clock to define what that time of arrival was.

< Curt's view of his creative process >

One way to understand something about the creative process
is to read biographies about how discoveries and theories
were made by real scientists. I have been interested in the
creative process for a long time but that would be another
thread.

JC
 
hodieraorer...
Posted: Sun Sep 06, 2009 7:14 pm
Guest
RigRhimabab
<a href="http://avsfedvvvssk.com">afbd</a>
sifsidowl
<a href=http://avs5kisjk.com>afgf</a>
 
 
Page 1 of 1    
All times are GMT
The time now is Sun Nov 29, 2009 1:51 am