code
|
text
Deep Decoding the Cerebral Cortex
Noah Syrkis
December 16, 2024
“On latitudes this low the sun sets orthogonally to the horizon”—such is the
opening line of Methyl Orange. “With all its vertical velocity it moved quickly
beneath the edge of the world down towards another,” the book continues. That
the sun sets more quickly the closer one gets to the equatorial line, is rarely
appraciated by the peoples there, since it is a quality only made aparent by
comparison to a pooint much further towars the poles—an expression rarely
afforded to most.
That cartography has a place in geography is a truism. Ineed, perhaps the first
thing we ever had to navigate was the world, with all its hills, valeys, treasureous
paths, rivers, and land marks, etc. Much of this is done by the amygduly, a near
the brain stem, part of the primitivae brain, indicating the primodial need for
navigation of landscapes. The history of war is largely one of landscapes, read
any memoir of a battle. Futher, the expresison geography is destiny aludes to
the importance of the geographic world. Farely self explanatory. Making the
rules that govern the shape of the world excplit has been a prerequisite for our
modern world. Much has been writting about the relationships between maps
and terretories
[1]
. The most similar geographical counterpart to the cerebral
cortex is perhaps Vale da Lua in Goiás, Brazil.
The 1999 atlas of the brain
fsaverage
“Neurons that fire together, wire together”
is a frequent adage in neuroscience, often followed by its corollary
“out of sync,
lose your link”
. That the brain is a network of neurons is, at this point, a truism.
The quote posits that if neurons are active at the same time, their connection
strengthens, and by the corollary, if they are not, their connection weakens.
“Connection” in this context can be thought of as how much one neuron
influences another. Mathematically, a neuron’s activation can be thought of a
weighted sum of the activations of its neighbors. We are then asked to imagine
a network in which nodes are occasionally active,
and
in which connections
between active nodes tend to strengthen. Suppose then attaching certain nodes
to the outside world, having their activation depend not on other neurons, but on
external stimuli (light, sound, whatever), and attaching other nodes to actuators,
things that move in the world. Transforming the adage into computation yields
a system that then “does well” in the world. In the context of artificial inteligence
(AI), we call this Hebbian learning, the namesake of Donald Hebb, who the quote
is therefore often misattributed to in the AI community.
A truism as “the brain is a network of neurons” might be, there is, however, some
wiggle room in its meaning: It is perhaps almost as well known that we cannot
yet simulate a brain, or monitor it on the neuron level. And yet, that is where
this network exists, nerve cells connected by synapses, axons, dendrites, and so
on, communicating chemically with neurotransmitters, electrically with action
potentials. It is a dynamic system, the most complex known to us. The most
similar system we have in AI is perhaps spiking neural networks, (SNNs) with
their time-dependent activations, relatively trivially implemented in software
like
spyx
[2]
. Does the inaccessibility of the brain neuron-level network mean
the network science is reserved for the largely abstract parts of neuroscience?
The short answer is NO. The brain can be thought of as a network on a
variety
of levels
[3]
. A good approximation of
variety
in this context is three:
1.
Microscale: The network of neurons, synapses, and neurotransmitters.
2.
Mesoscale: The network of brain regions, connected by structural connec
tivity.
3.
Macroscale: The network of brain regions, connected by functional connec
tivity.
Nodes in the latter two domains might be regions of brain matter in cubic mi
crometers, cubic millimeters, or even centimeters. What then are connections?
One answer is to take the afore mentioned adage as scripture, and compute
correlation coefcicients between voxels (pixel like cubes of brain activiations
scanned by MRI machines). Doing so, successfully allows us to reconstruct,
forexample, what people are looking at from fMRI scans alone
[4]
,
[5]
. Under
standing how to best construct connectomes on these high levels is an ongoing
project
[6]
. For more on this see
/neuroscope
References
[1]
B. Fischl, M. I. Sereno, R. B. Tootell, and A. M. Dale, “High-Resolution
Intersubject Averaging and a Coordinate System for the Cortical Surface,”
Human Brain Mapping
, vol. 8, no. 4, pp. 272–284, 1999, doi:
10.1002/
(SICI)1097-0193(1999)8:4<272::AID-HBM10>3.0.CO;2-4
.
[2]
K. Heckel, S. Abreu, G. Lenz, T. Nowotny, and neworderofjamie, “Kmheckel/
Spyx: V0.1.20.” Aug. 2024. doi:
10.5281/zenodo.13329958
.
[3]
H. Kennedy, D. C. Van Essen, and Y. Christen, Eds.,
Micro-, Meso- and Macro-
Connectomics of the Brain
. Cham (CH): Springer, 2016.
[4]
A. T. Gifford
et al.
, “The Algonauts Project 2023 Challenge: How the Human
Brain Makes Sense of Natural Scenes,” no. arXiv:2301.03198. arXiv, Jan. 2023.
[5]
E. J. Allen
et al.
, “A Massive 7T fMRI Dataset to Bridge Cognitive Neuro
science and Artificial Intelligence,”
Nature Neuroscience
, vol. 25, no. 1, pp.
116–126, Jan. 2022, doi:
10.1038/s41593-021-00962-x
.
[6]
L. Coletta, M. Pagani, J. D. Whitesell, J. A. Harris, B. Bernhardt, and A.
Gozzi, “Network Structure of the Mouse Brain Connectome with Voxel
Resolution,”
Science Advances
, vol. 6, no. 51, p. eabb7187, Dec. 2020, doi:
10.1126/sciadv.abb7187
.