Semantics and Syntactic Representations

Harry Erwin erwin at trwacs.fp.trw.com
Tue Nov 30 11:53:34 EST 1993


Bressler, Coppola, and Nakamura, 'Episodic multiregional cortical
coherence at multiple frequencies during visual task performance,' Nature,
366:153-156, 11 November 1993, presents some interesting evidence on the
role of spatial synchronization in solving the binding problem. This paper
is part of mounting evidence that semantically meaningful patterns are
represented in the mammalian brain as spatially synchronized oscillations
in the ionic currents associated with K0 systems in the brain. At the same
time, Walter J. Freeman has recently presented evidence that his KIII
model of the olfactory system generalizes to all the sensory modalities.
This KIII model is characterized by a lack of invariance--the point at
which the initial semantic classification of a pattern as being of
interest or not of interest is at the initial entry of the sensory stimuli
into the brain. Simulation work has begun to clarify the nature of this
processing. It seems to involve rapid pattern matching coupled with
conditioning (to set up patterns for future matching) and habituation (to
ignore patterns in the input stimuli that are not of interest). The deeper
cortices appear to control these processes through feedback (reafference),
in a manner that appears to involve mechanisms that allow the deeper
cortices to condition earlier cortices to detect specific patterns of
sensory data (perhaps through pathways that allow the deeper cortices to
generate 'synthetic sensory data'). Since these reafferent pathways appear
to involve some sort of inverse morphism to that which translates the
afferent sensory data as it is processed, this allows the deeper cortices
to determine the semantically meaningful sensory patterns they are seeking
and to prime the initial sensory processing to respond with semantically
meaningful signals when those patterns are detected.

The detection mechanism is of some interest. It is a variant on
'stochastic resonance.' Feedback loops in the first few layers of sensory
processing generate chaotic signals that are added to/subtracted from the
sensory data to generate a chaotically modulated pattern for comparison to
the patterns 'of interest' and 'not of interest'. The signal is modulated
in those areas where no match has yet occured and stabilized in those
areas where a match has been declared. This process continues until the
pattern is identified. If this continues too long, the pattern is regarded
as novel and worth investigating by the deeper cortices. In any case, the
output of the sensory processing upon detection of a known pattern is a
set of spatially synchronized oscillations. It appears that the invariant
pattern of sensory data can be reconstructed from the efferent signal of
the sensory processing, although the lack of invariance seen makes it a
hard question how that is done. 

The portion of the efferent signal from the initial sensory processing
that carries the invariant pattern of sensory data can be identified with
'qualia'. 

Each cortex appears to control the previous cortex by inhibiting or
releasing the processing by which a chaotic signal is generated for
specific patches. This suggests that heuristic dynamic programming is a
major unifying concept on the afferent side as well as on the efferent
side. This hints that the 'executive' function will be found to be
self-modifying. That is, the reafferent control used to modulate the
chaotic resonance process will be found to be a portion of the efferent
signal generated by the executive. 

Cheers,
-- 
Harry Erwin
Internet: herwin at cs.gmu.edu (not erwin at trwacs.fp.trw.com!)
Working on Freeman nets....



More information about the Neur-sci mailing list