==========================

—CORTICAL COLUMNS—

==========================

 

 

[ CORTICAL COLUMNS ]

 

Now that we’ve worked our way from the gross nervous system into the brain, into the cortex, and down to the lowest level of our exploration – the neuron, we’ll start work our way upward to the next level – the cortical column.

 

[ COLUMN DESCRIPTION ]

 

1. The cortex is divided into columns or to confuse us all, cortical mini-columns or micro-columns. 

2. The diameter of a column is about 28–40 µm.

3. Cells in 50 µm minicolumn all have the same receptive field; adjacent minicolumns may have different fields 

4. Each column has roughly 100 neurons each – with a variation in those numbers by half, and double in the visual cortex.

5. There are 200,000,000 of these columns or minicolumns,

6. Columns are organized into groups called cortical modules, functional modules or hypercolumns depending upon the whim of the authors. 

7. There are Roughly 1,000,000–2,000,000 cortical modules.

8. Columns, cortical modules, and groups of modules do vary a quite bit, but the overall design is the same across the cortex.

11. Nerve fibers (axons) teriminate in one or more minicolumns, and nerve fibers from the same region terminate in the same module, but they are not otherwise ordered – they  have to learn their relation to one another.

9. Cortical modules, and cortical Columns like neurons within them perform the same common function across the cortex. In the literature this is referred to as the common cortical algorithm. So the HOW is the same across the cortex.

10. The WHAT – or WHAT they do – is determined by what nerve fibers they’re attached to, and what part of the body they’re attched to – and they adapt to the inputs they receive, and the problems that they solve.

 

 

 

[ ORIGINS OF LAYERS ]

 

1. The origin of the six or so major layers is an evolutionary artifact. The cortex is an outgrowth of the region of the hippocampus that creates a model – a model creatures ‘feel’, of their eyes, head, body, and limbs in relation to space – think of it as your awareness of your body and where all its parts are in relation to all others. This region, which we will explore later, doubled, folded over, and grew larger and larger until at some point the stem cells that determined the size of the neocortex mutated and tripled. So as in all things, our ability to remember, predict, imagine, think, and reason as we understand it, is an evolutionary extension of our knowledge of our body in space.  AND this knowledge of cortical ancestry is going to help us understand the function of these columns.

 

 

[ CORICAL COLUMN LAYERS ]

 

Layer I – This layer is called the molecular layer. As you can see from the diagram, it has very few neurons and cells. Instead, it is mainly composed of dendrites and axons that extend from lower levels of the neocortex.

 

across the whole cerebral cortex, Layer I  receives substantial input from specialized thalamus cells, again, maintaining attention during what appears to be ‘association’

 

Layer I is rarely discussed in the literature. I’ll discuss this briefly later, but my understanding is that it preserves attention during the process of subsequent higher level cortical association. 

 

Layer II- Because this is one of the outer layers, and it is composed of granule cells, sGranule cells receive inputs (mossy fibers) from the spinal cord and brainstem. These receive inputs from other areas of the neocortex. Think of it as an intra  association layer.

 

Layer III-  This is the external pyramidal cell layer. Pyramidal cells make up this layer; they are called “pyramidal” because their somas are triangular shaped. They Receive input from other cortical regions and output to other cortical columns. 

 

layers II and III can be combined. They serve as the cortical output layer – the Voting, Competing, or Producing Layer.

 

Layer IV- Located deeper within the neocortex, and composed of granule cells. Granule cells in this layer receive sensory input and relay it to adjacent neocortex columns. Layer IV is very thick in primary sensory cortex (i.e. the primary visual cortex).

 

Layer IV is the information input layer.

 

Layer V- Another layer of pyramidal cells makes up the neocortex. The cells in this internal pyramidal cell layer are larger than those in Layer III. Layer V is involved heavily in making motor movements.

 

Layer V is the Motor Output Layer

 

Layer VI- Many different types of cells make up this layer. Its structure isn’t very homogeneous, so it was called the multiform layer. it’s Function is Receiving and integrating information from the brain stem and outputs to the thalamus.

 

Layer VI is the Position in relation to the body layer. (I think that’s the best way to describe it – the local GPS receiver maybe?)

 

 

[BRAILLE ANALOGY]

 

Let’s see if we can find an analogy that helps us visualize all of this.

 

This is a book published in braille.

 

If you run your fingers over it you will find bumps and divots and blank spaces. THink of these as textures or shapes.

 

Patterns of These bumps and divots represent characters.

 

With practice, as you move your fingers across each block of the texture, you can identify characters.

Layer Four.

 

Over time, Characters accumulate in your mind as words. 

Layer 3 accumulates predictions of greater context.

 

Words accumulate into sentences, and lets pretend a sentence is the same as an object or model.

Layer 2 for objects, layer 6 for actions

 

And you unconsciously tell your hand what direction to keep moving to continue receiving the information.

Layer 5 sends instructions to the thalamus and the motor neurons.

 

And you associate the meaning of the sentence with the rest of your short and long term memory

Layer 3 Competes for attention with a vote

 

ANd other columns compete with their votes

first nearby then across the cortex. Trying to fit features, objects, entities, together into scenes, at whatever resolution (meaning distance) we are focusing our attention.

 

Now. Imagine that instead of something very clear like braille on a page, that your brain is much closer to an octopus, with many many sensors connecting to many columns, in many modules, predicting many many fragments and competing for objects that fit together in a scene where your attention is directed to the close or far, and intensely or relaxed.

 

 

 

[ COLUMN LAYER PROCESS OVERVIEW ]

 

Now let’s cover the most basic process, of how a column calculates.

 

DIAGRAM 1

 

On the left you see a column with color coded layers. 

 

DIAGRAM 2

 

Next the column layers abstracted. 

 

DIAGRAM 3

 

Next we see layer four. 

 

Layer four takes the input from the nervous system. As I understand it, the majority of these inputs radiate out of the thalamus, however nearby connections also enter and exit in layer four. the exception being your sense of smell which is uniquely directly connected (and therefore difficult to regulate.)

 

Layer four learns to predict a sequence of sensory motor stimuli.

 

So that’s our first sort of surprise. instead of storing a set of stimuli, we store a prediction from sets of similar stimuli.

 

If any of you have used photoshop – an application that works in pixels, versus say illustrator that works in vectors, our brains work like vectors, not pixels or photographs. We remember the curve of a line, not the points in the line.

 

So if we repeatedly sense a sequence of points that are close to that line or curve or shape or texture, or color we predict that line, or curve, or shape, or texture, or color.

 

 

[ PREDICTION ]

 

 

 

Prediction begins all the way down in the neuron.

 

Let’s assume 1000 synapses per neuron 

 

Let’s assume about 8-15 will generate a spike, meaning release an action potential and fire a neuron.

 

Let’s assume  firing about 20 synapses will create a robust estimation of a unique prediction.

 

That’s 8.2*10^59 or 825928413359200443640727373872992573951185652339949568000000

individual predictions.

 

Now, I don’t know if you know much about large numbers, but in operational terms that’s pretty much infinity right there. Meaning, you can’t live anywhere near long enough to use up those permutations.

 

Now, in one column, there are a hundred neurons, some number of which are in layer four.

 

When one cell fires, it tends to recruit only about 1% of nearby cells. So it only contributes to other cells firings. So that means cells work together in even greater combinations to produce predictions.

 

Again, because big numbers are meaningless, for all intents and purposes a single column can identify an infinite set of sequences. 

 

And that means it, and the one’s near it, can recognize the edge of a table, the curves of a cup or glass, and the different textures of ceramic, metal, wood, skin or fur.

 

SO, our goal is to recall that small sparse combinations generate accurate information i

 

[ BACK TO LAYER FOUR ]

 

So in that layer four rectangle, you can see a bunch of dots

Let’s pretend that those are synapses that can fire, some sequence of those synapses fire causing some sequence of neurons to fire, 

 

DIAGRAM 4

 

So let’s look at the next diagram, where two things happen.

 

Immediately that information is transferred up to layers two and three. where just as we predicted a line, curve, shape texture or color – we’ll say fragment – we start predicting fragments into features, and features into objects. In other words we accumulate series of fragments accumulating in features and and objects.

 

And over time we get better and better at predicting what is familiar from less and less information.

 

And if we’re uncertain, then we don’t disambiguate features and this is interpreted by the cortex as ‘uncertainty’.

 

And we can process many, many sensations in parallel, within the neurons, within the columns,

And the modules can operate in parallel

And the whole cortex can operate in parallel

 

But we are also going to add something new here, and that’s the blue line – the position relative to your body, head, and eye. The fragment, feature, object, scene’s position relative to you – where you are the baseline of measurement for everything.

 

This information comes from a little region on the edge of your hypothalamus that we are going to spend quite a bit of time on in our video on orientation, position, space, and location. For now, I’m  going to ask you to recall the hexagonal sphere from the previous videos and just imagine that another part of  your brain is calculating and feeding you the information necessary to now the position of the sensation, and the object it’s disambiguating.

 

DIAGRAM 5

 

In the next diagram we see the blue arrow sending the prediction down to layer six, which will perform that calculation using what’s called a grid cell layer. And it means what it sounds like – it calculates where stimulation exists in space in relation to the body, and accumulates it there.

 

This helps us understand that the top of the column tries to solve the problem of WHAT, and the bottom the problem of WHERE.

 

DIAGRAM 6

 

In the next diagram we see all of this information, again, continuously updated in a continuous stream of continuing increase in precision – very vast, and in parallel, where the more experience, and the more training you have in a sequence of stimuli the lower the effort and information needed to predict an outcome for ever sparser (less) information. 

 

This continues as the column learns the entire object, learns it’s place in space and suggests how to coordinate the related motor nervous system – if applicable. 

 

 

OUTPUT

Layers 2/3 offer a prediction (cast a vote) both locally and to the opposite hemisphere.

 

Layer 5 tells the thalamus that it has a prediction worthy of attention.

 

And layer 5 sends motor actions to be relayed to the brain stem and spinal cord.

 

[ ATTENTION ]

 

Attention functions as the reward for successful prediction, and closing the thalamic circuit allows neural connections to form, strengthen and grow.

 

We will cover the Thalamus and attention shortly. 

 

 

[ COLUMN AND MODULE COMPETITION ]

 

Our brains break problems into massively parallel units of work

Then aggregate the solutions to those problems.

And do so by competition.

 

Competition speeds disambiguation.

 

So for example, one column might predict a tennis ball and a soda can. Another a soda can and a cup and so on. Two can gang up on one so to speak and eliminate one of its candidates. This competition continues and the survivor gets the attention (meaning survives) for the next iteration.  

 

 

[ SENSORY INTEGRATION ]

 

Prediction Across the regions of the cortex with different senses, where different sets of sensory-motor predictions compete, find commensurability or not, and survive or not. 

 

 

 

[ NETWORKS ]

 

And all of these predictions are always and everywhere continuously competing in real time.

 

 

[ DECREASE OUR LEVEL OF RESOLUTION ]

 

Now lets decrease our level of resolution a bit and review the cortical algorithm in more general terms.

 

 

COLUMNS CONVERT SENSORS TO MEASURES

 

Every sensory cell serves as a measuring device.

Every nerve produces a stimuli but all stimuli are routhly the same: on-off.

It’s the origin of the stimuli combined with its on off activity that provides information.

Neurons convert these stimuli into SEMANTICALLY meaningful measurements by searching for patterns that survive competition over time.

Competition includes not less than coherence (overlap) with nearby cells performing similar measurements, AND coherence over repeated sequences.

By way of all of this competition of sequences within a narrow window of time and space, millions of independent stimuli are converted into measurements, and measurements into objects upon which we can focus our attention and gain even more information.

 

Do you see the geometry now?

 

 

FEATURE INTEGRATION

 

The process of iterative association and consolidation is called Feature integration model. It states that that one of the main functions of visual attention is to bind visual features – such as color, texture, shape, orientation, and direction of motion – together into coherent objects.

 

 

BIASED COMPETITION MODEL

 

This process of competing predictions is called The biased competition model. According to this model, numerous sensory or cognitive representations are active in the brain at any given time, but the brain’s computational resources allow only a limited number of representations to proceed through stages of processing. 

 

So, the various representations are always competing against each other for access to neural resources. In this competitive environment, attention is a mechanism that can bias selected representations for more elaborated processing.

 

 

[ VISUAL CONTEXT – GEOMETRY ]

So let’s recall our Geometry.

Vitruvian Man

Our Relations

Our Commensurable Positions 

Our Commensurable World

and our experience of the world reconstructed.

 

 

 

[ COLUMN SUMMARY ]

 

Now we can summarize.

 

1. The column is the basic processing unit of the neocortex.

2. Columns disambiguate stimuli into predictions of sequences of stimuli.

3. Columns remember correct predictions of sequences, not individual sensations like a camera.

4. Remembering Sequences adds memory of timing, and timing necessary for coordination.

5. Columns receive orientation and positional information via the thalamus, to provide commensurable in space.

6. Columnar layers divide the labor of producing predictions of changes in stimuli in time and space.

7. Neurons in columns compete to produce a prediction of a relations, fragments, features, then object.

8. Columns accumulate predictions to produce a prediction (meaning theory) of an object or set of candidate objects

9. Columns in a module compete or vote to produce a prediction of an object. 

10. Modules pass information forward for further integration with the information from other modules in a vast parallel process.

11. All of this occurs very fast, in massive parallel, in iterations, accumulating, holding, and discarding information 

12. The success of each prediction, is determined by completion of a thalamic circuit over time.

 

 

[ DIMENSIONS ]

 

And finally, Let’s cover our dimensions. unlike neurons, There aren’t many to cover with cortical columns and modules .

 

We use three 

 

1. PHYSICAL

Neuron (measure) > Column (fragment) > Layer (object) > Module (model) > Region-Area (integration)) > Network ( composition ) > Cortex (simulation)

 

2. INFORMATIONAL

Sensation > Sequence (time) > fragment (space) > feature > object > model (category) > entity (instance) 

 

3. LOGIC 

Competition for repetition (sequence), coherence (integration, overlap), attention (survival) 

 

 

 

 

our next lesson covers  orientation position, space and location

 

this is curt doolittle for the prop….

 

 

 

====

 

To another axon terminal or bouton 

is called Axosynaptic

 

Releasing only into the extracellular fluid – yes they can do that.

is called axo extra cellular

 

Directly to the bloodstream

is called axo sec retory

 

 

—

 

So the takaway is that  The reason  axons  branch into many axon termials, boutons, and syapses is largely to guarantee sufficient supply of neurotransmitter release to meet demand. There are also undoubtably instances of some logical component to these connections, but primarily, the boutons can only process so much material at one time, and over so much time, so increasing the number of connections increase the supply of neurotransmitters.