By way of a preliminary, we extend the use of high values and low values. High values already have a role in column objects. We now add to that by saying that the defining values of patterns may not be either the high value or the low value. And that low value might be interpreted as the null value.
We call the repeated instances of a pattern which make up a layer object its elements. In the case of a line or a loop the elements have a natural order, if not a natural starting position.
We also talk of our data structure, the layered data structure which we hypothesise to be held on a patch of cortex, maybe a few centimetres square.
In order that a column object could be qualified by a layer object, we allowed wild cards. That is to say we allowed the high value to function as a wild card, allowed a point on the grid taking the high value to match any pattern value, which option, for the sake of consistency with what will follow, we now restrict to interior pattern values.
We now broaden our definition of a pattern, without having much idea of whether these new patterns will be very amenable to neural computation. We now say that our patterns are defined by the values of the perimeter, values which may be neither the high value nor the low value. Note that a pattern will only have an interior in the case that it has more than two rows and more than two columns.
So the two patterns on the left in the figure that follows do not have interiors, while the two on the right do, with the green pattern taking null values for the interior.
We then build contiguous patterns on our grid up into the elements of our layer objects in the same way as before, with the first rule being that there have to be at least two elements in a layer object, so that we know we have a pattern in the absence of any kind of declaration, the sort of thing which one does with data in regular computer programs, but which neurons might have trouble understanding.
And another rule being that the pattern defining a layer object is the smallest such pattern. So in the example which follows, it is twelve copies of the small rectangle which define the twelve elements of the layer object, not three copies of the larger rectangle.
This device of soft centred patterns allows us to include more data in our layer object, and in particular to vary the content of a layer object across its elements. So as well as telling us that we have some serious data, something more than a bit of noise, the exterior of the pattern can also label the interior, tell us what function the pattern has in the host object. In a sense, we are using the exterior/interior of a pattern to function rather as a preposition/object or label/object in a sentence in a natural language, or, for that matter, in html.
This gives us some economy in the use of layers, allows us to do in one layer what might otherwise have required several.
A pattern of the old sort might be used to record the colour or texture of an object that is seen or remembered and which has been mapped onto one or more layer objects in our data structure, while a pattern of this new sort can be used to record the variation in colour or texture across that object.
Noting in these examples that the green shading cannot map directly onto the real world object concerned, not be quite retinotopic, as it were, because we have the space taken up by the yellow perimeters of the elements of the objects. Perhaps this, along with the tiled, slightly jagged edges of our layer objects, is something that can be ironed away by integration into the field which generates our consciousness of those objects.
The exterior of such a pattern is also a bit like a loop in the sense of reference 3, albeit restricted in shape to a rectangle and with one loop allowed to touch another. Rather different, we might find a use for lines defined by soft centred patterns, for lines which have variation along their length.
Lines, arrays and sequences
Combining such lines with the soft centres allows us to define a sequence using just the one layer.
The data, or at least most of it, can be held in the soft centre, rather than being on some adjacent layer, linked by position.
In this particular case, it seems unlikely that all the information shown above would be available to consciousness. There may just be a line of rather vague faces or there may be focus on just one of them. Perhaps as much as two adjacent.
The next example is rather different, providing something by way of a bridge between analogue and digital data, with this line of three objects being read as ‘AXE’. With the white cells of the three interiors taking the low or null value. A scheme which supposes that each letter has already been mapped to one element – a process of separation which can it itself take a fair bit of computing.
Alternatively, the elements of the object might contain words rather than letters.
Alternatively again, there might be some other way of coding the elements. The words or letters might, for example, be coded as sound bites. Or as integers – which would imply a map being held somewhere, somehow, to allow one to move between the integers and what they stood for.
Other devices
More speculative and more elaborate still, we might have the following.
The mainly red perimeters tell us that we have some data about the host object, data that has been computed, imposed top-down rather than seen or heard. The two by two square at the bottom left of each element marks what lies to the right as meta data, further specifying what the contents above are all about. The first element might be some estimate of the weight of the host object, the second the price and the third its emotional content. Is this a good object or a bad object? Am I proud or ashamed of it? While the white rectangles of numbers (random in this case) represent the data itself.
The figure that follows shows how this data might be bound to the layer object, the large blue rectangle, to which it belongs. Blue on one layer marking out the space occupied by our object, red on another carrying the descriptive data, not present in space in the same way at all, bound by virtue of occupying the same portion of space, on the underlying grid.
In this context, the tricky question is how patterns of data of this sort get into consciousness. Patterns of colour is one thing, patterns of data of this sort is quite another.
Another speculation would be to generalise the line to a small array, perhaps insisting for ease of identification that the rows and columns all line up exactly, rather than approximately in the way of the green and yellow line above.
Summary of the possible applications for soft centred patterns
Colour
Variations of colour, texture or sensation across a layer object. We might expect the mapping from brain colour to world colour to be continuous. But we do not expect taste or smell to vary in space in this sort of way.
Sound
While something heard, perhaps bit of music, might vary in time rather than space. Maybe a layer object made up of a number of elements of a soft centred pattern. With the soft centres holding the sound. A number of elements holding the recent past, a rather smaller number holding predictions that had been made for the most recent part of the recent past and an even smaller number holding predictions for the future.
Position
Some objects might have a position and a shape in space which are not a simple consequence of their position on our grid. Soft centred patterns might be used to make the mapping. A mapping which we might expect to be locally continuous – but which might be complicated in other ways.
And thinking of continuity, worth noting that we might expect nearness relations in the outside world, in the stuff coming in from the various senses, to be preserved in their expression here, in this inside world. Distortions yes, tears no – at least most of the time.
Sequences
Letters of a word.
Words of a sentence – written.
Words of a sentence or a phrase – spoken.
A sequence of objects, perhaps just objects of interest for one reason or another, perhaps the constituent steps of some task or program.
Classification
Economical storage of classification data about a layer object. Age, sex, price and weight sort of thing, stuff which is not obvious from the externals. Maybe computed, maybe drawn down from memory, maybe a bit of both.
The classification data needed for consciousness might be in-situ, or might be elsewhere, in which case it would be pointed to.
Such classification data might include data about the emotional content of an object.
Conclusions
In building with these soft centred patterns, we have generalised the notion of a pattern first set out at reference 1 to give us something which carries rather more information. Allowing us to code in mainly spatial terms, using geometry rather than syntax. Or put another way with our syntax being geometry. Speculative in that, while all this is motivated by the need to build a bridge between layers of neurons in the cortex and data of the sort we – or a computer might – understand, we have no idea how easily neural networks on computer, never mind real neurons in real brains, might tackle such things.
PS: with thanks to the owners of the faces. Picked up from google, but they are probably real faces of real people for all that.
References
Reference 1: http://psmv3.blogspot.co.uk/2016/12/from-grids-to-objects.html. Introducing patterns on a grid.
Reference 2: http://psmv3.blogspot.co.uk/2017/01/layers-and-columns.html. A post which develops the structure of layers and columns which we are proposing to host the experience of consciousness. A rather long post with six diagrams and ten references.
Reference 3: http://psmv3.blogspot.co.uk/2017/01/lines.html. Making lines and loops out of patterns.
Group search key: sra.
No comments:
Post a Comment