What have neural networks achieved?

[Richard Granger]

Michael Arbib wrote:
>>  So: I would like to see responses of the form:
>>  "Models A and B have shown the role of brain regions C and D in functions E
>>  and F - see specific references G and H".

Models of the induction and expression (storage and retrieval) mechanisms
of synaptic long-term potentiation (LTP), the actual biological change in
connection strength that underlies at least some forms of real
telencephalic memory in humans, have led to novel hypotheses of the
functions LTP gives rise to in the actual circuitries in which it occurs.

One instance is the olfactory bulb-cortex system:  LTP in this system
(which was shown by Kanter and Haberly, Brain Research, 525:175-179, 1990;
and Jung et al., Synapse, 6: 279-283, 1990), endogenously induced by the 5
Hz "theta" rhythm that occurs during exploration and learning (Komisaruk,
J.Comp.Physiol.Psychol., 70: 482-492, 1970; Macrides, Behav.Biol., 14:
295-308, 1975; Otto et al., Hippocampus, 1991), was shown in models to lead
to an unexpected function of not just remembering odors but organizing
those memories hierarchically and producing successively finer-grained
recognition of an odor over iterative (theta) cycles of operation
(Ambros-Ingerson et al., Science, 247: 1344-1348, 1990; Kilborn et al.,
J.Cog.Neurosci., 8: 338-353, 1996).

This is an instance in which modeling of physiological activity in
anatomical circuitry gave rise to an operation that was unexpected from
behavioral studies, and had been little-studied in the related
psychological and behavioral literatures.


>>  The real interest comes when claims appear to conflict.

Other studies of the olfactory system have yielded quite different
predictions; this raises the question of whether animals cluster and
subcluster odors behaviorally, and whether paleocortical cells respond
selectively to different sampling cycles of clusters of similar odors.
These important issues are far from resolved; some relevant experimental
evidence on behavioral learning of odors is found in (Granger et al.,
Psychol. Sci., 2: 116-118, 1991); and on unit cell activity in cortex
during learning in (McCollum et al., J.Cog.Neurosci., 3: 293-299, 1991; and
see Granger & Lynch, Curr.Biol., 1: 209-214, 1991, for a review).


>>  What about the role of hippocampus in both spatial navigation and
>>  consolidation of short term memory?

Many studies begin with observed behaviors linked to medial temporal
regions by lesion studies and some chronic recording; it has been pointed
out that the connection specifically to hippocampus, as opposed to
surrounding perirhinal cortical regions, is difficult.  Moreover, the range
of behaviors, from spatial navigation to short-term memories, are
suggestive of emergent operations arising from combinations of more
fundamental mechanisms that may be occurring within the various modules of
these brain areas.

Not only is the medial temporal region composed of hippocampus, subiculum,
and overlying cortical structures, but these naming conventions occlude the
richness of circuitries within.  The "hippocampus" is composed of three
extraordinarily distinct structures (dentate, CA3 and CA1), each of which
consists of very different cell types, synaptic connections and local
circuits, and which are strongly connected with neighboring structures
(subiculum, pre- and parasubiculum, and superficial and deep entorhinal
cortex).  Of interest from a modeling point of view are the distinct
functions that emerge from the physiological operations of these disparate
circuits as well as composite functions arising from interactions among
them.  It will be interesting to uncover not only how these circuits
participate in well-studied behavioral circumstances such as navigation and
memory consolidation, but also what heretofore unexpected functions may be
found to arise from their action and interaction (Lynch & Granger,
J.Cog.Neurosci., 4: 189-199, 1992; Granger et al., Hippocampus, 6:
567-578).

It's worth mentioning that a special issue of the journal "Hippocampus"
dedicated to "computational models of hippocampal function in memory"
appeared as Volume 6, number 6 (1996); it may be a useful reference for
this part of the discussion.

- Rick Granger