Skip to main content

Brain Microarchitecture : Feedback from Higher-order areas to Lower-order areas

Some questions that arise in Machine Learning involve the prospect of using feedback from Higher-order areas (downstream) to Lower-order areas (upstream), and using Global Knowledge for Local Processing.  A desire to gain insight into those issues from Neuroscience ("how does the brain do it?") led me to some fascinating investigations into the Microcircuits of the Cerebral Cortex.  This blog entry is a broad review of the field, in the context of the original motivating questions from Machine Learning.

Starting out with a quote from the “bible of Neuroscience”:

From Principles of Neural Science, 5th edn  (Online book location 1435.3 / 5867).  Emphasis and note added by me:
Sensory pathways are not exclusively serial; in each functional pathway higher-order areas project back to the lower-order areas from which they receive input. In this way neurons in higher-order areas, sensitive to the global pattern of sensory input, can modulate the activity of neurons in lower-order areas that are sensitive to local detail.
For example, top-down signals originating in the inferotemporal cortex might help neurons in V1* to resolve a detail in a part of the face.
(*) V1 is the Primary Visual cortex

Thalamus : passageway to the Brain Cortex

Thalamus means “anteroom”.  It's a brain region that features prominently in the brain circuits described in later sections.

The thalamus comprises many projections to the cerebral cortex; hence the name “anteroom”.  Most of the information going to the cortex goes thru the Thalamus.  For example, the lateral geniculate nucleus, the main central connection for the optic nerve to the primary visual cortex in the occipital lobe, resides in the thalamus.

(Perhaps more familiar to many people is the “hypothalamus”, meaning “below the thalamus”, which is responsible for the central control of homeostasis in the body.)

Both the thalamus and hypothalamus are part of the Diencephalon, an inner part of the brain (considered by some, but not all, neuroscientists to be part of the brain stem)

The Cerebral Cortex

The mammalian cerebral cortex, the grey matter encapsulating the white matter, is composed of layers. The human cortex is between 2 and 3 mm thick.

The number of layers is the same in most mammals, but varies throughout the cortex. In the neocortex, 6 layers can be recognized although many regions lack one or more layers.

The neocortex is the newest part of the cerebral cortex to evolve (prefix neo meaning new); the other part is the allocortex, which has just 3 or 4 cell layers.
From the perspective of ML, my take is that the brain cortex first evolved in a smaller “search space” (with fewer cell layers), and then parts of it “colonized” a larger search space of evolutionary possibilities.  Akin to training a smaller network, and then adding extra layers and using the previously trained state as a starting point for the new training.

The Cortical Layers

Connections "up" and "down" within the thickness of the cortex are much denser than connections that spread from side to side.

Layer IV is the main target of thalamocortical afferents from the thalamus.

Layer VI sends efferent fibers to the thalamus, establishing a very precise reciprocal interconnection between the cortex and the thalamus. That is, layer VI neurons from one cortical column connect with thalamus neurons that provide input to the same cortical column. These connections are both excitatory and inhibitory.

Cortical microcircuits are grouped into cortical columns and minicolumns. It has been proposed that the minicolumns are the basic functional units of the cortex.  Functional properties of the cortex change abruptly between laterally adjacent points; however, they are continuous in the direction perpendicular to the surface. There is evidence of the presence of functionally distinct cortical columns in the visual cortex, auditory cortex, and associative cortex.

Studies mentioning “Top-down Signals”

A 2017 article gives experimental evidence (from monkeys with brain lesions) that:
“the prefrontal cortex (PFC) has long been considered a source of top-down signals that bias selection in early visual areas in favor of the attended features

Source:  Paneri, S., & Gregoriou, G. G. (2017). Top-Down Control of Visual Attention by the Prefrontal Cortex. Functional Specialization and Long-Range Interactions. Frontiers in neuroscience, 11, 545. doi:10.3389/fnins.2017.00545     link

Selected quotes from a 2007 review article:

All cortical and thalamic levels of sensory processing are subject to powerful top-down influences, the shaping of lower-level processes by more complex information.
The general idea of top-down influence is that complex information that is represented at higher stages of processing influences simpler processes occurring at antecedent stages. Whereas some of the earlier work on spatial attention—the most studied instance of top-down modulation—suggested that significant influences of attention are found only at high levels in the visual pathway, it is becoming increasingly clear that even at the earliest stages in cortical sensory processing the functional properties of neurons are subject to influences of attention, as well as other forms of top-down modulation.
The higher-order information may include learned, internal representations of the shapes of objects and of the abstract syntax of object relationships. It may also include information about behavioral context, which would include attention, expectation, and perceptual task.
These influences may not even be specific to cortex, but wherever one sees feedback connections, including thalamus. This study showed even stronger attentional effects in the LGN* than in V1/V2**. Top-down influences are not unexpected in the LGN since it receives input from many more V1 neurons, by orders of magnitude, than it receives from the retina.
(*) LGN : the Lateral Geniculate Nucleus is a relay center in the thalamus for the visual pathway
(**) V1 is the Primary Visual Cortex

Source: Gilbert & Sigman (2007),  “Brain States: Top-Down Influences in Sensory Processing.”  Neuron, Volume 54, Issue 5, 7 June 2007, Pages 677-696        link

From a 2016 article:

(1) “lower-order” visuotopically organized cortical areas, some of which receive their principal or a substantial, direct thalamic input from the dorsal lateral geniculate nuclei (LGNd), send numerous “feedforward” associational projections to the “higher-order” visual areas

(2) beyond the primary visual cortices, information about pattern/form vs. motion is processed along two largely parallel “quasi-hierarchical” feedforward streams

(3) higher-order areas send numerous associational “recurrent” or “feedback” projections back to lower-order areas

Source:  Huang JY, Wang C and Dreher B (2017)  Silencing “Top-Down” Cortical Signals Affects Spike-Responses of Neurons in Cat’s “Intermediate” Visual Cortex. Front. Neural Circuits 11:27. doi: 10.3389/fncir.2017.00027   link

Microcircuits of the Cerebral Cortex

From 2012 article “The Cell-Type Specific Cortical Microcircuit: Relating Structure and Activity in a Full-Scale Spiking Network Model”  (link)   Excitatory (black) and inhibitory (gray) connections with connection probabilities >0.04 are shown.

Central to the idea of a canonical microcircuit is the notion that a cortical column contains the circuitry necessary to perform requisite computations and that these circuits can be replicated with minor variations throughout the cortex.

George and Hawkins have suggested that the canonical microcircuit implements a form of Bayesian processing (George, D., and Hawkins, J. , 2009. Towards a mathematical theory of cortical micro-circuits. PLoS Comput. Biol.5, e1000532. )

The most popular scheme — for Bayesian filtering in neuronal circuits — is predictive coding (Srinivasan et al., 1982; Buchs-baum and Gottschalk, 1983; Rao and Ballard, 1999).  In this context, surprise corresponds (roughly) to prediction error.  In predictive coding, top-down predictions are compared with bottom-up sensory information to form a prediction error.  This prediction error is used to update higher-level representations, upon which top-down predictions are based.

To predict sensations, the brain must be equipped with a generative model of how its sensations are caused. Indeed, this led Geoffrey Hinton and colleagues to propose that the brain is an inference (Helmholtz) machine (Hinton and Zemel, 1994;Dayan et al., 1995).  A generative model describes how variables or causes in the environment conspire to produce sensory input. Generative models map from (hidden) causes to (sensory) consequences. Perception then corresponds to the inverse mapping from sensations to their causes.

In predictive coding, representations (or conditional expectations) generate top-down predictions to produce prediction errors. These prediction errors are then passed up the hierarchy  in the reverse direction, to update conditional expectations.

The generative model therefore maps from causes (e.g., concepts) to consequences (e.g., sensations), while its inversion corresponds to mapping from sensations to concepts or representations. This inversion corresponds to perceptual synthesis, in which the generative model is used to generate predictions. Note that this inversion implicitly resolves the binding problem by explaining multisensory cues with a single cause.

Diagram source:  Canonical Microcircuits for Predictive Coding (2012)

Note:  intrinsic connectivity = within a cortical column ;
extrinsic connections = between columns in different cortical areas

The above is a simplified schematic of the key intrinsic connections among excitatory (E) and inhibitory (I) populations in granular (L4), supragranular (L1/2/3), and infragranular (L5/6) layers. The excitatory interlaminar connections are based largely on Gilbert and Wiesel (1983).

Forward connections denote feedforward extrinsic corticocortical or thalamocortical afferents that are reciprocated by backward or feedback connections. Anatomical and functional data suggest that afferent input enters primarily into L4 and is conveyed to superficial layers L2/3 that are rich in pyramidal cells, which project forward to the next cortical area, forming a disynaptic route between thalamus and secondary cortical areas (Callaway, 1998).  Information from L2/3 is then sent to L5 and L6,which sends (intrinsic) feedback projections back to L4 (Usrey and Fitzpatrick, 1996). L5 cells originate feedback connections to earlier cortical areas as well as to the pulvinar, superior colliculus, and brain stem.

In summary, forward input is segregated by intrinsic connections into a superficial forward stream and a deep backward stream. In this schematic, we have juxtaposed densely interconnected excitatory and inhibitory populations within each layer.


Popular posts from this blog

Graph Databases (Neo4j) - a revolution in modeling the real world!

(UPDATED 11/2022) - I was "married" to Relational Databases for many years... and it was a good "relationship" full of love and productivity - but SOMETHING WAS MISSING! Let me backtrack.   In college, I got a hint of the "pre-relational database" days...  Mercifully, that was largely before my time, but  - primarily through a class - I got a taste of what the world was like before relational databases.  It's an understatement to say: YUCK! Gratitude for the power and convenience of Relational Databases and SQL - and relief at having narrowly averted life before it! - made me an instant mega-fan of that technology.  And for many years I held various jobs that, directly or indirectly, made use of MySQL and other relational databases - whether as a Database Administrator, Full-Stack Developer, Data Scientist, CTO or various other roles. But there were thorns in the otherwise happy relationship The root cause: THE REAL WORLD DOES NOT REALLY RESEMBLE THE

Life123 : Quantitative Modeling of Biological Systems

(UPDATED 8/2022) - Are we ready to embark on a next-generation detailed quantitative modeling of complex biological systems , including whole-cell simulations?  An anticipated up-jump in computing power may be imminent from Photonics computers (which I discuss here ), and GPU's are rapidly gaining power as well...  Are we in ready state to put existing - and upcoming - power to good use? This is a manifest, and a call to action What's Life123? It's about detailed quantitative modeling of biological systems in 1-D, 2-D and full 3-D, as well as a multi-faceted software platform for doing so. What's (pseudo-)1D?  For now, let's say it's like the inside of a long, thin tube - with no interactions with the tube.  Likewise, (pseudo-)2D can be thought of as a Petri dish, with no interactions with the lid or the bottom. Website : A purposeful decision to also utilize 1D and 2D But why?  Yes, it's in part about "walk before you run&quo

Discussing Neuroscience with ChatGPT

UPDATED Feb. 2023 - I'm excited by ChatGPT 's possibilities in terms of facilitating advanced learning .  For example, I got enlightening answers to questions that I had confronted when I first studied neuroscience.  The examples below are taken from a very recent session I had with ChatGPT (mid Jan. 2023.) Source: In case you're not familiar with ChatGPT, it is a very sophisticated "chatbot" - though, if you call it that way, it'll correct you!  'I am not a "chatbot", I am a language model, a sophisticated type of AI algorithm trained on vast amounts of text data to generate human-like text'. UPDATE:  this article focuses on some of the impressive abilities of ChatGPT.  For a good glimpse of its weaknesses, in the context of poor intuition about Physics, as well as Math errors, check out this great short video:  ChatGPT does Physics For a high-level explanation of how ChatGPT actually works -

D3 Visualization with Vue.js : a powerful alliance (when done right!)

[UPDATED MAY 2022]  D3.js is a very powerful visualization tool, especially for specialized/custom needs...  On the flip side, it's rather hard to use - with a steep learning curve. Even worse if one also wants interactivity ! But why is D3 so hard/clunky to use?  And what can be done about it? Spoiler alert: Vue.js (or other modern front-end framework) to the rescue - if done right... All code in the examples is available in this GitHub repository . The Root of the Problem In a nutshell, what makes D3 awkward to use is that, for historical reasons, it tries to do too much : most painfully, it uses an old way to do direct DOM manipulation (i.e. restructuring the page layout) - an operation that nowadays is superbly handled in a far more friendly way by modern front-end frameworks, such as Vue.js Document Object Model ( DOM ) is a programming interface for web documents.  In simple terms, it's the structure of the elements on a web page (text, images, etc.) Let the front-e

To Build or Not to Build One’s Own Desktop Computer?

“ VALENTINA ” [UPDATED JUNE 2021] - Whether you're a hobbyist, or someone who just needs a good desktop computer, or an IT professional who wants a wider breath of knowledge, or a gamer who needs a performant machine, you might have contemplated at some point whether to build your own desktop computer. If you're a hobbyist, I think it's a great project.  If you're an IT professional - especially a "coder" - I urge you to do it: in my opinion, a full-fledged Computer Scientist absolutely needs breath, ranging from the likes of Shannon's Information Theory and the Halting Problem - all the way down to how transistors work. And what about someone who just needs a good desktop computer?  A big maybe on that - but perhaps this blog entry will either help you, or scare you off for your own good! To build, or not to build, that is the question: Whether 'tis nobler in the mind to suffer The slings and arrows of OEM's cutting corners and limit

A "Seismic Shift" in Longevity Science : Mainstream Acceptance + Large Funding

"You are incredibly prescient!"   I woke up to those words from a former colleague on Jan. 19, 2022: the bombshell announcement that the Chief Science Officer of pharma giant GSK, where I worked until recently, will become the CEO at the new, $3 BILLION longevity science company Altos (presumably also funded by Amazon's Jeff Bezos.) Big Pharma is at long last embracing Longevity Science. The corollary: longevity science is entering Mainstream (with capital "M") But let me backtrack... The Decade of Longevity Science When Harvard professor David Sinclair declared the 2020's to be the " decade of the paradigm shift about age reversal ", one could perhaps be dismissive of it as just an outburst of enthusiasm... But in the past couple of years, we're seeing strong evidence that his forecast is right on the mark! While I worked at GlaxoSmithKline - a giant, top-10, pharma company - I vigorously advocated forming a Longevity Science dept., and sp

PET/CT Combined Scanners - a 2018 Breakthrough of the Year... and a Personal Story

Image source Recently, a co-worker in her 20's was diagnosed with a brain tumor!  At times like these, the importance of medical imaging jumps to the fore! Most people have heard of CT ("CAT") scanners – at least enough to know that they don't actually involve cats – but less well-known are PET scanners (which likewise don't involve pets!), and the synergistic combination of the two. A Marriage Made in Heaven What do those scanners do?  And why are they being combined in single devices? Voted 2018 Breakthrough of the Year by a science magazine , the improved PET/CT combined scanner has been a game changer. The EXPLORER PET/CT scanner – the world’s first medical imaging system that can capture a 3D image of the entire human body simultaneously – has produced its first human images. Developed by UC Davis scientists and a multi-institutional consortium, EXPLORER can scan up to 40 times faster, or use up to 40 times less radiation dose, than

RDF Triple Stores vs. Property Graphs : How to Attach Properties to Relationships

Time for the opening shot of a series about Semantic Technology , and in particular contrasting-and-comparing the opposing (but perhaps ultimately complementary) camps of:   RDF Triple Stores , aka Triples-Based Graphs.   For example, Blazegraph or Apache Jena   (Labeled) Property Graphs .  For example, Neo4j or Blazegraph (For this article, I'll assume that you have at least a passing acquaintance with both.  Here is background info on Triplestores and Property Graphs ) It’s my opinion that modeling in terms of Subject/Predicate/Object triples (aka RDF ) might be appealing to mathematicians or philosophers for its minimalist foundation (though a lot of baroque add-on’s quickly come out of the closet!) Modeling in terms of (Labeled) Property Graphs might be appealing to computer scientists, because such graphs appear more usable and less clunky once you start actually doing something with them. Perhaps because I straddle both the Math and CS camps, I’m currently on t

Anti-Aging Research: Science, not Hype

Last updated December 2022 Q: "How is aging a disease?" A: "It's a dynamic system that veers away from its homeostasis (normal equilibrium point): hence a form of slow-progressing illness. Labeling it as 'natural' is a surrender to our traditional state of ignorance and powerlessness, which fortunately is beginning to be changed!" The above is my standard answer to an oft-asked question. The science of aging is by all evidence very misunderstood by the general public.  Hype, misinformation and unquestioned assumptions often prevail, unfortunately. Aging as a systemic breakdown of the body, rather than a series of isolated events and conditions. This 2013 diagram from NIH is a good way to jump-start contemplating the big picture: The diagram originates from the Cell journal: The Hallmarks of Aging   Telomere shortening is perhaps the one most talked about - but just one of several processes.  As stated in the above paper: Each