If you could
record from 10-20 synapses at the input end of a neuron,
and display all the spike trains arriving to it through them,
you would most of the time see something like this
in other words
the spike trains are usually random and independent.
But then every once in a while something interesting happens, as I show here at the bottom,
where for a few brief moments, some of these usually independent spike trains are clearly and sharply correlated.
These repeated volleys are well distinguishable from the usual random firing
and the neurons can detect them too, because the volleys, if they are prominent enough, will make blips in the membrane potential.
According to my theory,
the "Legéndy theory,"
the meaningful signals in the brain
are carried by these volleys;
and the usual random firing,
like what’s at the top part of the slide,
is just useless noise.
I have to emphasize this
the statement goes against the ruling school of thought among most theoreticians at this time,
which bases all descriptions of the brain on the average spike rates of the neurons,
treated as functions of time.
and my point is
that dealing only with the average spike rates
amounts to dealing only with these upper tracks
and completely missing the occasional volleys in the lower tracks, and with them missing most of the meaningful communication in the brain.
This makes a huge difference in theoretical methodology,
because once the analysis is re-focused on these volleys, and other similar surprising events,
the usual computer simulations of evolving time-dependent spike rates become useless;
and instead one has to concentrate on how these volleys are generated,
what information is put into them,
and whether they can reach the neurons they are intended for.
But the real cool part of it is
that the revised rules
which I develop in my book,
are sort of fun to work with, because the issues take the form of problems in logistics,
in which the essential materials being moved around are the fragments of locally available information.
the challenge being
that no piece of tissue can act on information that has no way of getting to it.
let me next show you
the new methodology in action,
and apply it to the old elusive problem
of visual shape processing.
first let me say a few words
networks generating the volleys.
These are known as cell assemblies,
which are interconnected neuron groups that sometimes become unstable and ignite,
and each time they do they send out one of these volleys.
They are usually made up of cells close together in the cortex, which is why when you record from two cells picked up by the same electrode
see something like this cross-correlogram
redrawn from Noda and Adey.
The sharp peak in the middle
shows the number of times beyond the random rate
where the two cells fire within one millisecond of each other.
Well, each of those
corresponds to the ignition
of a cell assembly where both of the recorded cells are members.
Now, let me show you an example of a cell assembly: the direction-coded cell assembly,
which is perfect for encoding the shapes of visual images.
The cells in this one are presumably in layer 2-3
of the primary visual cortex, V1,
all close together in one column,
so that the group
responds to one point-like node in visual space;
and they have axons branching
so that they can only reach the nodes displaced from that node
in one particular direction.
Now, let’s imagine a cell surrounded by the axonal branches of such a direction-coded cell assembly.
Here of course this same cell is also reached by similar groups of axons coming from other directions meaning that it can hear the firing from all these differently direction-coded cell assemblies.
What’s interesting is when the cell receives the usual random noise from most of these groups but receives a series of volleys from one of them.
Since all these cell groups carry their own chemical markers,
the neuron that receives this input is in a position to know
that the volleys
come from this direction.
The next step in the logic is to consider the signals received by a cell in the next visual area, V2, which receives axons from many such cell groups,
and imagine that the cell hears the following combination of firing:
In other words it hears that two oppositely direction-coded cell groups send out volleys in sync with one another.
the group protocols -- The group protocols
can be designed to make sure that this only happens when
the two cell groups are interconnected;
in other words synchrony implies that the
two cell groups in V1
have to be in each other’s axonal field, like this
otherwise the axons of one group would miss the cells of the other.
And what this in turn means
is that two opposing directional groups
mapping to two different retinal locations can broadcast the relative direction
of these retinal locations by joining up and synchronizing their volleys to each other.
The reason this is important to shape processing is
that the shape of any polygon can be described by describing the relative directions
of its corner points.
it is necessary to notice
that the directional cell groups by themselves cannot do this,
because synchronized ignitions can only compare two points at a time.
To transmit the shape of a polygon
the identities of the points must be preserved from one point pair to the next.
For instance in order to transmit the shape
of a triangle via relative directions,
signals must be added to express
which points are shared between which sides.
This can be done by introducing a new type of cell assembly,
made up of cells,
which I’ll call kernel cells,
with very short
so this kind of assembly
can only sync up with a directional cell assembly having cell bodies in the same cortical column with it.
Then to tell a cell in V2
that two directional groups in V1 have their cell bodies in the same column,
the kernel cells simply have to show that they can ignite together with both groups.
When they ignite together with these directional cells they show that they are in the same place with them;
and when they ignite together with these
they show that they are at the same place with them as well.
In this way the two directional co-ignitions can be joined on their shared point.
Now, this kind of joining
is a very powerful trick
because it can combine many
two element relations to make
a single many-element relation,
which is how a whole visual image
can be transmitted inside the brain
as a single Gestalt.
First let me show you how a triangle can be transmitted via synchronized ignitions.
It goes like this.
As you see,
this is a fairly busy drawing because it involves nine neuron pools: six pools of directional cells and three pools of kernel cells;
but the idea is simple:
the directional groups must co-ignite by conjugate pairs and the kernel groups
must co-ignite with both directional groups in their respective corners.
One may note that the ignitions
don’t have to be all ordered like this; they can all keep repeating on their own channels, independently and at the same time.
Now – by doing just more of the same thing,
the V1 cortex can equally well transmit a more complex shape,
say, something like this
it just has to approximate the shape by a polygon and cut it up into triangles
adding some redundancy -- adding some redundancy;
and go from there.
In complex polygons there are usually more than two directional lines meeting in one point;
for instance point number 3 here is the meeting point of five lines, and co-ignitions must be
arranged between five directional ignitions with the same set of kernel cells:
So – now you see what I meant when I said a moment ago
that a directional cell assembly could hold the key to encoding the shape of a visual image.
Of course here I only gave you a rough sketch of the reasoning;
there are many details to be filled in, for instance conveying the difference between visible contours and diagonals;
but you get the idea;
and the details are spelled out in my book.
if you’d like to take the challenge,
and would like to learn how to construct this kind of logical reasoning,
I encourage you to read my book,
which contains tons of examples;
so we can all make headway
in the effort to understand the circuits in the brain.