sleep

[Richard Long]

schultz at sedal.su.OZ.AU (Simon Schultz) wrote:
>In article <3vgqal$gs9 at ixnews7.ix.netcom.com>, janyoung at ix.netcom.com (Janis Young) writes:

>P.S. By coincidence, I just attended a very interesting and relevant seminar
>by Gardner-Medwin today. His theory is along the lines I suggested below,
>but storage occurs during REM (not slow-wave as I suggested below). He has
>a model which suggests computational reasons for two phases of sleep which
>correspond to slow-wave and REM sleep. These reasons involve something called
>"enriched recall", a mechanism to reduce {\em confusion} in an
>autoassociative memory. The idea is not to refresh by presenting the original
>memory, but by one modified slightly so as to reduce confusion. (Which is
>the same as generalisation, but bad, rather than good).
>

In such a system, it seems to me that the mechanism of "enriched recall" and
that of ordinary training of your autoassociative memory are the same.  If 
this is true, why don't we remember (the vast majority of) our dreams?  Also,
Why would enriched recall not seem to our autoassociative memories like other 
experiences?
 
>
>How about the following idea? That sleep is in fact an "active refresh"
>system for memory, analogous to the active refresh we need to apply to 
>dynamic RAM memory chips to extend the storage time associated with the physics
>of the storage medium. (Even though in the DRAM the memory is not associative, 
>the analogy still holds).
>
I think that the fact that in DRAM the memory is not associative, as you 
point out, means that the analogy does not hold.  The problem with 
associative memories is that there is no one-to-one correspondence between 
a memory and a synapse.  Furthermore, I would argue that the level of 
active information processing that accompanies memory storage and retrieval
is such that no clear distinction can be made between which synaptic 
process contributes to which memory.  How would you know what to refresh?
If you refresh a synapse that shouldn't be refreshed, I would expect you
would degrade the system.

>Assume for a moment that the fundamental cortical long-term storage mechanism
>is associative long-term potentiation (LTP). The physics of this mechanism
>appears to dictate a time for complete decay of a stored synaptic
>enhancement of a few weeks. This means that we should refresh all important
>memories at least every few days, preferably once per night. Assume that
>memories, both in hippocampus and neocortex are both stored in an autoassociator(error-correcting coder) of some kind [1]. Sharp-wave rhythm corresponding to
>communication back from hippocampus to neocortex is found in slow-wave sleep.
>This could correspond to the actual active-refresh memory consolidation
>mechanism. REM sleep would involve reloading the hippocampus with the
>important information learned during the preceding day. [2] Memory is thus
>cycled through a hippocampal-neocortical active refresh loop.
>
I would like to comment on this in two ways.  First, I like the idea of a
hippocampal-neocortical active refresh loop.  I've always thought that the
experimental evidence regarding the hippocampus and memory suggests that
the hippocampus is a kind of "resonance chamber" for keeping certain
temporal and frontal cortical circuits activated, possibly needed for the
chaining of cortical attractors to form an attractor cascade which might
correspond to event memories and consciousness.  It might be true that
sleep in involved in some sort of hippocampal-neocortical dynamics.
However, still have still have a problem of what refresh means.  It might be
that the hippocampus' role is more one-to-one, and so would admit of a
refresh mechanism, which is then consolidated into the cortex's more
complicated and interwoven system of "memory" slowly over time.  But then
we still have the problem of how this process can affect our memories
without "affecting" them :-) (i.e. without leaving a trace that it occured).

>Note: this does not preclude different "daytime" function of the hippocampus
>in episodic memory etc. Possibly the hippocampal episodic memory system
>would be involved itself in working out what are the important memories
>from the day that need to be refreshed. Projections from amygdala etc. would
>of course help in this.
>
Experimental evidence is on your side here.  hippocampal activity is strongly
value and emotion laden.  You may be on to something :-)

>Actually, I've no idea how memories are stored neocortically, but whatever
>the circuit is, I think its guaranteed to need refresh of some kind. I
>suspect that it may be computationally more desirable to use a refreshed
>moderate-term storage mechanism (discard the rubbish) than to evolve an
>indefinte-duration storage device.

Yes, but discarding the rubbish could not likely happen on the synaptic
level, since it is impossible to decide which individual synapses 
constitute rubbish in a complex dynamcal system such as the brain.

>
Richard Long
long at acme.ist.ucf.edu
University of Central Florida