From newshub.ccs.yorku.ca!torn!utcsri!rpi!usc!cs.utexas.edu!uunet!trwacs!erwin Thu Jul  9 16:20:30 EDT 1992
Article 6420 of comp.ai.philosophy:
Xref: newshub.ccs.yorku.ca comp.ai.philosophy:6420 comp.ai.neural-nets:3449 sci.cognitive:119
Path: newshub.ccs.yorku.ca!torn!utcsri!rpi!usc!cs.utexas.edu!uunet!trwacs!erwin
>From: erwin@trwacs.fp.trw.com (Harry Erwin)
Newsgroups: comp.ai.philosophy,comp.ai.neural-nets,sci.cognitive
Subject: Nonlinear Distributed Memory
Keywords: memory holography cerebellum
Message-ID: <654@trwacs.fp.trw.com>
Date: 6 Jul 92 11:45:56 GMT
Followup-To: sci.cognitive
Organization: TRW Systems Division, Fairfax VA
Lines: 27

One of the comments I received on my "Generalized Distributive Memory"
posting was that holography is linear and one desired characteristic of
associative memory is that it be non-linear. An consequence of a linear
memory model is that it _reproduces_ the memorized pattern, rather than
finds new patterns, thus it is not creative.

I'm not sure I want non-linear memory, particularly in the cerebellum,
since my goal is the reproduction of a pattern of motor impulses, not the
creation of a new pattern. The motor cortex is responsible for devising
new patterns, the cerebellum is only responsible for memorizing and
replaying those that the cerebellum has devised, although the nature of
the "memorization and replaying" process _automatically_ supports
generalization as well. In any case, non-linearity in the brain is
_localized_ at the axon hillock, which suggests non-linearity is only
associated with long-range neuronal interactions, even with interactions
between distinct regions of the brain. The activities within the dendritic
arbor--the dendro-dendritic interactions--both chemical and electrical
synapses--appear to be linear, although here we have to remember that the
data are encoded in the inter-pulse intervals (not frequency, and only in
the first few interpulse intervals in a pulse train, since the train
converges rapidly to a limit cycle).

Cheers,
-- 
Harry Erwin
Internet: erwin@trwacs.fp.trw.com



