An alternate view of memory functions


#1

As a connectionist a make a clear distention between a detector NN and
a generator NN.

Detection NN would be one layer of the cortex, for incoming data.
Generator NN for mid part and later layer of the cortex for generating memories and
outputs.
That is regenerating a binary activation bit. A real magic trick.

Detector to detect food. Generator to remember the flavor, generator to make
the motors grab it.

A “latent value” is used to get data from Generator NN in machine learning.
But the Thalamus generate Latent values to address memory location and
turn on motors??? Or dose it even have a number system???

I think something like chain of paper punch cards.
Jacquard loom:

So I say that pulse spike moving through chain of never is used as a Latent value.
The pulse generates also generate a sequence of images and motor movements.
The chain is in the thalamus come out and feed into the cortex.

When a person pick up a book and start reading from a random location or wake up in strange place after a wild party the detection NN comes into play.
Detection NN hit places in the daisy chain where features of the surrounding
environment are recorded. Many detection, from the local environment, will cause a
spike to come into existence and to keep the spark alive. Continues detections must
occur to keep the spike alive.

Once a Spike has become stable, anywhere within a the very long daisy
or in a section of many broken long chains, then consciousness comes into existence
within this system!

The detection NN and generator NN team are a search light looking at the green
coloring around the lette “L”, like looking at the world through a straw, or a pinpoint
scanner, in a book. The spike of consciousness is here and can use eye motor
to complete a pattern loop of daisy chain of commands. Or jump to a new pattern loop.
It can keep using motor eye commands to search out the edges of the next letter
and generate and a built up sequence of letters to identify where it is in novel
of Mobby Dick. How exited this brain is. But emotions are of another section
of the brain, and for another time and place of discussion.

But what is said here is consciousness that a primitive non self aware animal
can achieve. Yes there is still allot more.

Law of NN:
NN convert input data to output data.
For the generator NN,
Latent spikes are, the input data, transformed into motor command and expected
images.

For the detector NN, detection of certain objects or data are concerted into
binary activation of one. When they a present right here and now and zero when they
are not.


Numenta turns attention to The Thalamus!
#2

Are you saying that Neural Network mechanisms exist in the brain? As in, trained via backpropagation/gradient descent?


#3

The detection NN does not use back propagation. It uses unsupervised learning.
in computer simulating of a real brain no, it not the star algorithm for detecting data.
But if by chance two rare SDR come into existence.
it can be used to fine a point in between that has
similarity with the outside data point. Bur this is rare event that two SDR bit come
int existence as neighbors and are data similar, and close.

In the generator NN back propagation, gradient of decent would work quiet well. But i
do not know if a real brain uses this method???


#4

I’m certain it doesn’t, I was just trying to understand what you were proposing.

Usually connectionists are researching ways to produce intelligent behaviour purely from deep learning (mathematical) techniques, but you’re weaving biology into the mix too which is why I was confused.


#5

@keghn_feem
I saw this on twitter and thought of your descriptions:

https://dataspace.princeton.edu/jspui/handle/88435/dsp01nz8062179