TAPe models the workings of the mechanisms of perception_post_25

Company

Company

TAPe models the workings of the mechanisms of perception

TAPe represents an important link between biology, cognitive and other sciences that study the structure and functioning of the brain, and new computer solutions, including AI.

01

02

Neurobiologists claim that photoreceptors send the signal to the neural retina. Inside the retina, there are four layers of nerve cells which process the visual signal. Then the output cells of the retina form the optic nerve with the help of their axons. About a million axons go further into the brain. Generally, from the point of view of the perception system, the eye is a part of the brain, not a separate independent organ.

04

Data collected by neuroscientists suggest that the basic objects recognizable by the brain are a dot and a line. A line, a 12 degree turn, a total of 15 lines — that is 180 degrees. And these lines, according to neuroscientists, are innately recognised. There are some patterns, filters, properties within the human eye, or the visual system, that are imprinted in this way, and the human visual system tries to detect them in any object.

Let's consider AlexNet visualization of the filters for one of the initial convolutional networks, which has had a big impact on computer vision algorithms. Why are the filters the way they are (in shape, number, features, etc.)? One could say that it's just the result of random search by the AI. Or you could say that even the lines found by AI models are similar to those recreated by the human brain.

07

05

But is there an understandable hypothesis of what happens to these lines and dots afterwards? Why are they exactly like that? Why is there this exact number of them, why couldn't there be 10 at 18 degrees for example? And what happens to these lines, or rather to the data that the brain receives from an object through the visual system?

06

They are somehow — how? — recognised, some lines are detected in a particular object — but why exactly that many lines, not more or less? — and what happens to them next? How and why do they combine in the brain into this very object and not another? What laws are at work here?

The "mathematical" or, more precisely, computational framework of TAPe forms parts of a general system, which we call the Language of Thought system after Jerry Fodor. We see the Language of Thought as a mechanism more habitual to the brain's functioning than traditional mathematics or AI methods.

10

09

Apparently, these innate primary elements, intuitively discovered by human genius, were the basis for the creation of visual images that are used even in natural human language. And the "mathematical" framework of the TAPe might be revealing the laws by which visual images are formed in the brain. "Mathematical" is put in quotes because TAPe does not involve the usual math — differentials, integrals or back propagation of error, as well as Bayes' theorem.

08

Human vision is directly connected to the brain being the main provider of information about the world around us — this is a known fact of neurobiology. "'Innate' line detectors in the visual perceptual system are the primary elements found by evolution that are necessary and sufficient to recognise the most complex visual patterns."

03

This argument, among other things, led to the conclusion that the picture "digitized" by the retina has about 1 million pixels. But the main question is how this information, be it in the form of pixels, is processed? However, according to TAPe, the very existence of "pixels" as elements of perception is highly questionable, because it is clearly an unnecessary, uninformative, energy-consuming "division" of reality into some patterns subsequently used by the brain. Pixels are an extreme simplification of biology for the needs of technology.