🚀 go-pugleaf

[Please note follow-ups]

In article <343DEDFE.8F6B8872@qi3.com>, Steven Packham
<packhste@qi3.com> wrote:

>Louis Savain wrote:
>
>> 
>>   I've acknowledged the importance of work in operant conditioning.
>> This work however sheds no light on the actual mechanism at the single
>> neuron level.  Without a knowledge of this mechanism, no progress can
>> be made.
>
>This is  a very interesting point. What is your question here? What
>mechanism are you searching for one of activation or one of learning
>or modification of synaptic strengths, or one of temporal coding or
>information relay.

  All of the above with the possible exception of the "temporal
coding" part.  Temporal coding may mean different things to different
people.
 
>> > But a  temporal  relation
>> >alone  is not sufficient (as Kamin showed long ago). If one  goes
>> >beyond  "delay",  to "trace" conditioning, it looks  like  higher
>> >levels of processing such as the hippocampus is required - but as
>> >I've said before, this is already getting into levels of learning
>> >in the neuroaxis...
>> 
>>   I'm sorry, this is nonsense.  A signal is but message indicating
>> that something happened at a specific time.  The neuron processing the
>> signal has no idea what that something is and could not care less.  It
>
>It has some idea that it is inhibitory or not.

  Good point.  The YES (excitatory) and NO (inhibitory) forms of
synaptic connections are what we (Marengo Media) call "extractors."
One should look at a signal or event as having both a beginning and an
end.  A YES synapse means that the target neuron is "interested" in
whether or not the event has started whereas a NO synapse means that
the neuron is interested in "knowing" whether or not the signal has
ended.  Note that these are strictly temporal considerations which say
nothing about the nature of the sender.

> And what of the length of
>the proteins which are capable of being formed. Is there no message 
>capability here.

  Biochemical processes are relevant to a level of abstraction which
is below that of signal processing.  We are computer technologists and
we are interested in emergent intelligence for control purposes and in
complex signal processing.  What is important to us are signals and
what happens to them before and after they are processed by a neuron.
Certainly, whether the signal is handled chemically or electrically is
of interest to us but this interest stems mainly from intellectual
curiosity.  From a practical standpoint, we are more interested in
knowing that we can simulate an analog of the process on a digital
computer.

>> only gets the signal and that is it.  To make sense of anything it
>> needs to find several signals to determine their relationships, if
>
>specifically it must have two or more signal or the ability two have
>two or more signals. the absense of one signal could be enough to
>provide contrasting information to the other present signal as long as
>there are times when there is more than just the one signal. In other
>words two possible active inputs or more inhibitory or excit.

  I agree.

>> any.  Temporal relationship is the **only** type of relationship that
>
>	You argue for the existance of a temporal relationship and 
>add that a signal is a message that something happened at a specific
>time. what codes the time message is the question? David proposed to
>coding methods being 1 delay and 2 a higher cognitive function. You 
>claim neither but I fail to see the proposed coding technique.

  [You and Bill Modlin seem to have a knack for asking the most
pertinent questions and I appreciate it.  It shows a good grasp of the
subject and also an acute interest.]  David Longley suggested that
there is something else besides temporal relationship involved in
event processing and I strongly disagree.  When it comes to events,
temporal relationship is *it*.  There is absolutely nothing else that
can be learned from discrete signals, higher cognitive functions
notwithstanding.  Temporal relationship comes in three categories, a)
before, b) after, and c) contiguous.  In our model, 'before' and
'after' are dealt with at the single neuron level and contiguity is
handled by the network's parallelism.  I regret that I can't say too
much more about it until our package is ready for publication. 

>[...]
> You seem to imply that perhaps David might suggest a higher
>funcitioning of Intellegence such as attention in the search light
>idea or something and prod on that what ever higher intelligente feature
>David might suggest is superstition.

  I don't know what David was suggesting but whatever it is, I am
convinced it is "intensional superstition" stemming from "folk
psychological idioms" (some of David's favorite terminology being
thrown back at him :-)) because, from the standpoint of neural event
processing, there is no logical basis for the introduction of any form
of relationship between events other than temporal relationship.

> That there is no other organized 
>unit in the brain capable to guide learning which can be named other
>than temporal. Very well for your definition of temporal. which you
>just stated as simply being events, or signals with the word temporal
>tacked on.

  Correct.  Once you accepted this temporal/causal paradigm, the
search for possible candidates to serve as fundamental processing
units for intelligence becomes mightily constrained.  In one fell
swoop, an enormous number of research avenues can be eliminated from
consideration.  For examples, my company and I no longer have any
interest in pursuing approaches to AI having to do with formal logic,
symbol manipulation, knowledge representation, deduction, restricted
domain problem solving, production rules, Bayesian networks, belief
systems, conventional ANNs, frames, scripts, and a bunch of others.
[I'm not attacking anyone doing research in those fields, just
pointing out that we are no longer interested]  This, IMHO, is a
breakthrough, even though many will not agree for whatever reasons.
But it does not matter.  What matters is whether or not it will lead
to practical results.

>[...]
>
>Are you just searching for a learning mechanism in the neuron?

  Yes and we think we found an excellent one.

>A very basic one that can act over several or one neuron connections
>distance to provok learning?

  I'm not sure what this means.
 
>[...]
>> 
>>                Temporal relationship
>>                is the only type of
>>                relationship that can
>>                exist between events.
>
>between events by definition. Events denotes time.
>As defined by you Temporal events are times.

  Agreed.  And event is slightly more than just non-descript times
however.  It consists of a beginning time and an ending time.  BTW,
throughout my comments on this subject some of you will notice my
frequent use of complementary pairs of concepts such as
beginning/ending, before/after, YES/NO, weak/strong, etc.
Complementarity is a significant part of our philosophy.  Another
equally significant principle is simplicity.

>>   Therefore, intelligence and learning must be about temporal
>> relationships.  Nothing else.  Deny at your own risk.
>
>This suggest the impossibility to divide time into quanta.
>where is the time exactly now.

  I'm not entirely sure what you mean here.  Let me say that our model
makes no assumption about time per se and we do not attempt to
represent time internally.  Our model deals only the relationships.
Since an event processing unit (neuron) has a temporal resolution
(about 1 millisecond in natural neurons) one could loosely say that 1
millisecond is the time quantum.  But I don't see the real importance
or advantage in doing that.

>Where one calculation can be performed on what exists presently.
>At the instant where two neighboring neurons receive signal of contrast
>which interact at an intersection point of another neuron and the
>resultant effect they have on that neurons in turn activation. 

  I don't think I understand what this means.

>Can it then be claimed that this can't happen in any form and any
>small enough time measure to be considered a singular event for 
>those three neurons? And if can happen for three why not four?
>Where to place the limit on intelligence and learning being temporal
>(time separated events) 
>relationships instead of a mix of temporal and positional or modeling
>or mappings or structuring or attention directing.

  I am not sure what your point is here.  Any system will have a
temporal resolution beyond which its temporal processing can't learn
anything.  We humans have this problem but we learn enough to create
powerful instruments, such high-speed cameras, to compensate for our
handicaps.

>Are the events the time between signalling?
>
>Or are events signalling with coded time stamps?
>
>Or are events just two signals into one neuron?

  An event consists of a beginning temporal marker and an ending
temporal marker.

>IT would  be very interesting to now about the complete temporal coding
>theory.

  Ah.  Note that David Longley has been clamoring for us to release
our promised package, and we will.  We have received messages from a
few interested and concerned parties alerting us to possible conflicts
with their own published work and even to possible patent
infringement.  We are conducting a thorough search of the litterature
on the subject to make sure that we will not be stepping on anyone's
toes.  This is taking longer than expected but it pays to be prudent.
The extent of the amount of research work that has been and is being
conducted in this area has taken us by surprise.  Although we may be
wrong (unlike Longley we are inductive learning animals) we think we
have something unique which is worthy of publication and we believe it
can only get better once others start contributing to it.  Please bear
with us.

Regards,

Louis Savain

President, Marengo Media, Inc.