Hierarchical Brain

Abstraction and prediction-enhanced selection

This is my name for the second of four levels of description of the afferent processing of sense and other data. It is level 3 of 7 in my hierarchical structure, “above” memory-enhanced coincidence detection and lateral inhibition but “below” symbol schemas.

This page describes the three components of abstraction and prediction-enhanced selection and how they emerge from the three components of memory-enhanced coincidence detection and lateral inhibition, although it is not a one-to-one transmutation.

Abstraction

• Memory-enhanced coincidence detection, a major part of level 2 processing, carried out recursively over many hierarchical levels leads to emergent behaviour that can be called compression, extraction of invariance, or abstraction.
  • There is compression because at each hierarchical level of coincidence detection processing, one output signal represents two (or more) input signals. So over many levels of processing one output signal can end up representing many input signals.
  • There is abstraction because at each hierarchical level of coincidence detection, one output signal represents two (or more) small pieces of information about an object or concept. So over many levels of processing one output signal can end up representing a large-scale object or concept.
  • At each hierarchical level of afferent processing the information represented is more abstracted (it originates from a wider area, or has a bigger receptive field), more invariant (less changing) and more compressed (a smaller amount represents the same thing).
  • The end result of many levels of processing is many invariant representations. This is the beginnings of an understanding of the incoming data, which can be called intelligence.
  • The afferent processing examples give more details on how this can happen.
• Compression and abstraction of sense data happens when I sense something that I have sensed previously¹.
  • To put it another way, if I sense something I have never sensed before, there will be no compression or abstraction.
  • So for a new-born baby, everything is captured, nothing is discarded².
  • In this sense, a new-born baby has a better memory than they will ever have in their future life.
• The result of many levels of compression can be called abstraction, or extraction of invariance, or generalisation, in that the general features or average behaviour of a thing are extracted. This is how the meaning of a concept is extracted³.
  • Abstraction can be seen as a form of compression (see abstraction - compression) because compression of data extracts and keeps the important parts that are common between occurrences.
  • The Scottish philosopher and psychologist Kenneth Craik gave a description of abstraction in 1943 describing it as the separation of the common characteristic and relationships from any particular physical object or situation⁴.
• Jeff Hawkins’ Memory prediction framework uses the term abstraction in the same way that I do, meaning the processing and compression of sense data to create invariant (unchanging) representations.
• One particular well-documented example of abstraction is translation invariance⁵, which can also be described as the detection of translational symmetry.
  • A good example of this is how I detect that a frisbee is the same thing when it is viewed from a side-on angle as when it is viewed from above, or when it is coming straight at me, edge-ways on.
  • I have given simplified examples of how this might be done in afferent processing example 4 and example 6.
• Research has confirmed that the brain does process data in a hierarchical manner to extract common features and hence create more and more abstracted symbols, and that this has also been shown to work in artificial computer networks⁶.

Prediction

• Memory-enhanced coincidence detection also creates or strengthens efferent connections back towards where the data came in.
  • These can be used for prediction, because a circuit can be formed from higher up the hierarchy back towards the source of the incoming data.
  • The better the prediction, the more likely that circuit is to win the competition (see selection below).
  • The afferent processing examples give more details on how this can happen.
• As with abstraction and compression above, prediction can only happen when a thing has been sensed before, so the brain of a new-born baby performs no prediction².
  • When a young child sees a frisbee for the first time, there is no prediction, so all the sense data is saved, and this starts to set up efferent connections.
  • When the child sees the frisbee for a second time (even microseconds later), these efferent connections will be used, which is a low-level form of prediction, and a connection will be made to the already-existing representation of the original symbol (see afferent processing examples).

Selection

• Lateral inhibition, another component of level 2 processing, carried out over many hierarchical levels leads to the selection of the most important signals and the inhibition of others.
  • This selection process can be influenced by the strength of the input signals (bottom-up or afferent).
  • It can be also influenced by the strength of prediction signals (top-down) that use efferent connections.
  • This overall process of selection is sometimes described as biased competition and is a low-level description of how the high-level process of attention works.
• A balance is maintained between excitatory and inhibitory connections in all neurons⁷.
  • This means that there is competition between neurons to respond to any stimulus.
  • This is carried up the hierarchy so that there is also a competition between groups of neurons when they come to represent symbols (see model of my world for more details).
  • It is also a vital component of the architecture of the brain and the way my thoughts flow from one things to another.
  • This fractal architecture also leads to chaotic dynamics in the brain as a whole.

---

References For information on references, see structure of this website - references

1. ^ On Intelligence or On Intelligence - Jeff Hawkins with Sandra Blakeslee St. Martin’s Press USA 2004
   Page 82, talking about invariant representations: “I believe a similar abstraction of form is occurring throughout the cortex, in every region. This is a general property of the neocortex. Memories are stored in a form that captures the essence of a relationship, not the details of the moment. When you see, feel, or hear something, the cortex takes the detailed, highly specific input and converts it to an invariant form. It is the invariant form that is stored in memory, and it is the invariant form of each new input pattern that it gets compared to.”
2. ^ ^ How emotions are made - The secret life of the brain - Lisa Feldman Barrett 2017 Pan Books (UK) or see GoogleScholar.
   In the chapter entitled “How the brain makes emotions”, page 113, third paragraph: “The infant brain is missing most of the concepts that we have as adults. ... Not surprisingly, the infant brain does not predict well. A grown-up brain is dominated by prediction, but an infant brain is awash in prediction error. So babies must learn about the world from sensory input before their brains can model the world. This learning is a primary task of the infant brain. At first, much of the onslaught of sensory input is new to an infant’s brain, and its significance is undetermined, so little will be ignored. ... Infants absorb the sensory input around them and learn, learn, learn. The developmental psychologist Alison Gopnik describes babies as having a 'lantern' of attention that is exquisitely bright but diffuse. In contrast, your adult brain has a network to shut out information that might sidetrack your predictions, allowing you to do things like read this book without distraction. You have a built-in 'spotlight' of attention that illuminates some things, such as these words, while leaving other things in the dark. The infant brain’s 'lantern' cannot focus in this manner. As the months pass, if everything is working properly, the infant brain begins to predict more effectively. Sensations from the outside world have become concepts in the infant’s model of the world.”
3. ^ Concept cells: the building blocks of declarative memory functions - Quiroga 2012
   doi: 10.1038/nrn3251 downloadable here or see GoogleScholar.
   From beginning of “Conclusions and open questions”, page 595: “Our thoughts are based on abstractions and the attribution of meaning to what we sense or recall. Concept cells in the human hippocampus are the pinnacle of this abstraction process and provide a sparse, explicit and invariant representation of concepts, which, I have argued, are the building blocks for memory functions, such as the creation of associations, episodic memories and the flow of consciousness. This interpretation is supported by a large number of studies showing the role of the MTL in memory and the creation of associations, and also by the specific characteristics of concept cells discussed above.”
4. ^ The Nature of Explanation - Kenneth Craik Cambridge University Press 1943 or see GoogleScholar.
   Chapter 6 entitled “Some consequences of this hypothesis”, pages 69 and 71 under the heading “Abstraction and brain mechanisms”: “...one of the characteristics of memory and perceptions is the recognition of identity or of similarity. To recognise a thing is surely to react to it, internally, or overtly, as the 'same thing' to which we reacted on a previous occasion. ...the more subtle forms of recognition - recognition of similar shapes of different sizes, or casting their images on different parts of the retina, of the 'three-ness' of three apples or three oranges, or of relations such as 'to the left of'. These represent degrees of 'abstraction', that is, separation of the common characteristic from any particular physical object or situation, and of 'relation', that is, of position in space and time relatively to other objects.”
5. ^ The brain from inside out - Gyorgy Buzsaki 2019 Oxford University Press
   doi: 10.1093/oso/9780190905385.001.0001 or see GoogleScholar.
   Page 292, in Chapter 11 entitled “Gain and Abstraction”, under the heading “Abstraction by Different Reference Frames”: “It is remarkable that the transformation between eye-centered and body-centered coordinates requires only a neuronal gain mechanism. If such a simple mechanism work for one coordinate transformation, perhaps it can be deployed for similar purposes in other circuits. That is, the body-centered information carried by neurons that read the output of the parietal cortex can undergo further transformation at the next level of computation using the same type of gain control mechanism. Along the way, ever more complex features of the retinal information can be extracted. When an object is viewed, approached, and touched from different directions, the multiple experiences may strip off the particular conditions in which the object was sensed, leaving only its essential features, regardless of where they appear in the visual field. This is known as 'translation invariance', an abstraction.”
6. ^ The global landscape of cognition: hierarchical aggregation as an organizational principle of human cortical networks and functions - Taylor, Hobbs, Burroni and Siegelmann 2015
   doi: 10.1038/srep18112 downloadable here or see GoogleScholar.
   End of Abstract on page 1: “...we objectively sorted stratified landscapes of cognition, starting from grouped sensory inputs in parallel, progressing deeper into cortex. This exposed escalating amalgamation of function or abstraction with increasing network depth, globally. Nearly 500 new participants confirmed our results. In conclusion, data-driven analyses defined a hierarchically ordered connectome, revealing a related continuum of cognitive function. Progressive functional abstraction over network depth may be a fundamental feature of brains, and is observed in artificial networks.”
   End of discussion, page 16: “From tangible sensory inputs, symbolic content may progressively emerge as information is processed deeper into the brain’s structural network, starting with inputs, and expanding in abstraction or refinement, resulting in intangible or deep symbolic content in the structural pinnacle of the human brain network. These experiments successfully addressed a proverbial 'elephant in the room' an assumption about brain function, sometimes taken as axiomatic, though at times objected, that seems simultaneously too grand, too simplistic, and too complicated to demonstrate comprehensively.”
7. ^ Daily Oscillation of the Excitation-Inhibition Balance in Visual Cortical Circuits - Bridi, Zong, Min, Luo, Tran, Qiu, Severin, Zhang, Wang, Zhu, He and Kirkwood 2020
   doi: 10.1016/j.neuron.2019.11.011 downloadable GoogleScholar.
   Summary, page 621: “A balance between synaptic excitation and inhibition (E/I balance) maintained within a narrow window is widely regarded to be crucial for cortical processing. In line with this idea, the E/I balance is reportedly comparable across neighboring neurons, behavioral states, and developmental stages and altered in many neurological disorders. Motivated by these ideas, we examined whether synaptic inhibition changes over the 24-h day to compensate for the well-documented sleep-dependent changes in synaptic excitation. We found that, in pyramidal cells of visual and prefrontal cortices and hippocampal CA1, synaptic inhibition also changes over the 24-h light/dark cycle but, surprisingly, in the opposite direction of synaptic excitation. Inhibition is upregulated in the visual cortex during the light phase in a sleep-dependent manner. In the visual cortex, these changes in the E/I balance occurred in feedback, but not feedforward, circuits. These observations open new and interesting questions on the function and regulation of the E/I balance.”

---

Go back to where you last came from or return to the top level summary.

These web pages have no adverts and no cookies, no data is collected or processed so there is no privacy policy, and access is not secure. You are free to use any material on these pages provided it is credited to this site and referenced to the correct page.

The views, opinions and statements in these web pages are my own personal opinions. I am an amateur neuroscientist with no formal training, and I have no accounts on any social media platform. Any comments, corrections or criticisms are welcome, although I don’t guarantee to reply - email me at info@hierarchicalbrain.com.

Page last uploaded Mon Feb 19 10:43:53 2024 MST