Congradulations on having your picture and work mentioned in the
Scientist. I was disappointed that your most valuable contribution was
not mentioned. Namely the finding that the reported taste doubles for
sweet, sour, bitter and salt is reduced by 1/2, when half of the tongue's
sensory system is knocked out.
As you know, I view these facts as confirmation of a wavelet and
correlational opponent-process which links all senses under a common
frame of reference. When the opponent wavelet information is knocked out
for half the data, the strength of the reported signal would double.
The reduction of the signal of salt by 1/2 is easy for everyone
to explain that are not in the COP theory model, but very difficult
to explain in my model. If information is overwritten on a reference
frequency and Na+ slowed down the reference frequency by raising
the voltage of the action potential, or increased the resistance then
a reduction of 1/2 would make sense in the COP model.
As you know Gilbert reported an effect of colors on odors, and you
had suggested phenomenon similar to this for taste. This seems to
me then to support an interacting global/local wavelet model.
So any taste would be like a gaussian wavelet oscillon. Your report on
6-n-propylthiouracil and its effect on taster and non-tasters supports
a gaussian wavelet model, in my opinion.
I have enclosed an article on modular decomposition that relates to these
issues. Ron Blue
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>Ghahramani and Wolpert (1997) reports evidence that visual motor
learning occurs through modular decomposition. Two expert
neuro specialization areas created by learning would send
hierarchical gaussian mixtures to create generalizations or multiple
relationships. "This relationship results from the assumption that
each expert is responsible for an equal variance gaussian region
around its preferred starting location, which corresponds to its
The modulation model was significantly a better fit to describe the
observed behavior than a linear model. A linear constraint model
would predict a linear generalization pattern. This was not
confirmed by the data. The visumotor system has limited
generalization to novel events which suggests local receptive field
The experimental results "show that learning two new visumotor
mappings, whether represented as vectors or postures, at two
starting locations, leads to a smooth sigmoidal generalization at
The research may be interpreted to support Correlational Opponent-
Processing for the following reasons: gaussian receptive fields model
is supported, modulation interaction that is sigmoidal supports
wavelet interaction, simultaneous multiple learning supports global
interaction and local interaction, and visumotor areas are
functioning as activating and inhibitory centers.
All these events are a normal consequence of the wavelet nature of
neuro processing. The neuro structure is a global history of
previous and current environmental stimulations. Behavior is
never dependent upon a single neuron. This process is almost
identical to the formation of physical oscillons in a vibrating
system with two frequencies (Umbanhowar, Melo, and Swinney 1996).
Oscillons modulate and exist due to the unseen wavelet interactions
of the two frequencies and the history of the system. Oscillons are
the observable memory in a vibrating system. That memory is made up
of a positive particle phase oscillon and a negative particle phase
oscillon. This particle oscillon can be thought of as a figure and
the apparent noise oscillations around the oscillon as background.
Notice memory consists then of figure and ground, local and global,
longterm potentiation and longterm desentization, short-term
potentiation and short-term desentization with all modulating in
time. Memory then is dependent upon reference frequencies, stimulus
overwrite on that frequency from a sensory field, correlational
opponent filters, oscillating oscillons created by interaction
wavelets by using neurotransmitters and evoked potentials. This
models the quantum dilemma of particle and wave at the same time.
Vannucci and Corradi (1997) at the University of Kent at Canterbury,
UK have written an interesting paper that relates to these issues.
The paper concerns wavelet shrinkage technique through orthogonal and
linear wavelet transforms, which allows decomposition of noisy data
into a set of wavelet coefficients so that noise can be removed by
shrinking the coefficients. NTC - Neutronics Technologies
Corporation's CORE processor uses similar methods to form wavelets, oscillons
and reduce noise by simply dividing the information into oppositional
halves. One half of a stimulus history interacts with current input
data which forms a new history. Stable oscillons and wavelets of
memory form from this interaction.
The Bayesian model is a summation statistical model with a mean
of zero with gaussian high and low bypass filters for wavelet
extraction. This also describes the CORE processor. The mother
wavelet generated by this Bayesian model suggest why Ricci, a
NTC robot can have self control and self directed behavior. Mother
wavelets would represent from a philosophical point of view an idea
or correspondence to a schema in the environment. Visual symbolic
representation of this is suggested by eigenfunction pictures at
http://www.neutronicstechcorp.com under the Technical Tangents
Covarience structure of random wavelet coefficients allows learning
and creativity to occur. The reason being that any mother wavelet
must have harmonic wavelets of lower strength. This supports
Ghahramani and Wolpert's (1997) conclusions and observations of
generalization and modulation in visumotor learning.
Additional research is suggested by Vannucci and Corradi report
of using BayesShink on blocks, bumps, heavisine and Dopler
signals seeded with gaussian white noise. The BayesShink is
successful in recovering the data with the exception of the Dopler
signal. The Dopler signal is distorted at the beginning
of the signal. This wavelet interpretation of neuro processing
should demonstrate problems then in a Dopler signal and allow
a way to find problems in the model.
Blue, R & Blue, W. (1996). Correlational Opponent Processing: A
Unifying Principle. available by email at
Ghahramani, Z. & Wolpert,D. (1997, March 27 ). Modular Decomposition
in Visumotor Learning. Nature pg 392-395.
Vannucci, M. and Corradi, F (1997, May). Some findings on the
covariance structure of wavelet coefficients: Theory and models
in a Bayesian perspective. unpublished report UKC/IMLS/97/05
Umbanhowar, Paul B.; Melo, Francisco and Swinney, Harry L. (1996,
August 29). Localized excitations in a vertically vibrated granular
layer. Nature p793-796.