Page 114 - AN-3-3

Advanced Neurology Evaluating plausibility of thalamic model

In artificial neural networks, Hebbian learning integrate cortical inputs to support higher-order cognitive
rules are typically used in pattern associator models, processes. For example, the thalamic matrix plays a
where a weight matrix W transforms a set of input crucial role in modulating cortical states and facilitating

 
)
(
2
1
m
vectors I = I , I , …I into a set of output vectors consciousness, as demonstrated by stimulation studies
that can recover consciousness following anesthesia.
18
 

)
(
p
2
1
O = O , O , …O . However, there is a limitation to In addition, the thalamus sustains cortical activity and
this learning rule in this application context. Unless the modulates neural synchrony, which is essential for effective
19
input patterns are orthogonal to each other, it is generally information transfer between cortical regions. However,
impossible to avoid contamination or “interference” many aspects of thalamocortical interactions remain
between the correct output for one input pattern and the poorly understood, including the precise mechanisms by
output for other input patterns. 6 (pp90-93),7 (pp79-82) Both cases which the thalamus influences cortical plasticity and the
involve vectorial encoding, where a specific perception is detailed role of inhibitory circuits in the thalamic reticular
20
represented by its coordinates in an n-dimensional space. nucleus. Moreover, the comprehensive impact of thalamic
In vectorial encoding, the limitation of orthogonality in dysfunctions on neurodevelopmental and neuropsychiatric
17
Hebbian learning can be addressed by decomposing each disorders necessitates further investigation.
input pattern into its mutually orthogonal coordinates. This All thalamic nuclei receive reciprocal corticothalamic
process is equivalent to performing principal component projections from layer VI of the cortical region that they
(PC) analysis. This would enable its successful application innervate, with corresponding conduction velocities that
8
to each coordinate of the pattern rather than the pattern induce resonant electrical activity in the circuit. Llinás
21
itself. et al. proposed that these loops would form vertical
22
If this is true for any pattern associator system that uses functional regions, known as resonant columns, which
Hebbian rules, it would be necessary to find biological are of significant interest for analysis. The activity of these
mechanisms for orthogonalization that enable this columns could potentially explain perceptual coherence in
kind of learning process in the brain. We suggest that a the brain, with temporal coincidence (binding) occurring
9
central orthogonalizer would produce the necessary set through integration mechanisms in the apical dendrites of
of orthogonal axes, or PCs, to adequately represent each cortical pyramidal neurons that receive signals from the
23
specific input pattern, allowing for subsequent Hebbian ventrobasal and centrolateral thalamic nuclei.
learning processes. There are indications that this These columns can be classified into two primary types
preprocessing may occur in the computations of thalamic of transmission relays (Figure 1): (i) first-order relays,
networks, normalizing and completing the afferent which receive subcortical sensory and motor inputs along
information that will feed into higher-level networks. 8,10-12 with inputs from the reticular formation and project to
The thalamus serves as the primary entry pathway to primary areas of the cortex; and (ii) higher-order relays,
the cortex, transmitting sensory, motor, and autonomic which receive inputs from layer V of the cortex and
information from the cranial nerves, brainstem, cerebellum, the reticular formation, constituting a component of a
and spinal cord to specific areas of the cortex. 13,14 All cortico-thalamo-cortical or trans-thalamic pathway. This
afferent projections undergo processing in thalamic nuclei, pathway retransmits information already present in the
24
and even in olfaction, where there is no direct input from cortex, but from one area to another. In addition to these
sensory neurons to the thalamus, there are indications of two types of thalamo-cortical projections that connect
the involvement of the mediodorsal thalamic nuclei, which with layer IV, there is also the paralaminar projection,
communicate reciprocally with primary and secondary which communicates with layer I. This projection is
olfactory areas. Thalamocortical networks play a crucial excitatory to the apical tuft dendrites of pyramidal cells and
15
role in various sensory, cognitive, and motor functions, simultaneously provides input to inhibitory interneurons
25 (pp8-9;16)
facilitated by complex and reciprocal connections that also communicate with these dendrites. The
between the thalamus and the cortex. These networks branches of the pyramidal cells create internal circuits of
are fundamental for processes such as attention, arousal, widespread excitatory feedback, both for themselves and
and cognitive control, with thalamocortical neurons for other pyramidal cells, while also supplying input to
26
receiving information from ascending sensory pathways interneurons that propagate lateral inhibitory feedback.
and projecting to the cortex, while corticothalamic There is a certain regularity in the cytoarchitectonic
neurons provide modulatory feedback to the thalamus. 16,17 organization of thalamic nuclei. Despite the classification
Recent studies highlight the diverse functions of thalamic differences between first-order and higher-order relays,
nuclei, which not only relay sensory information but also there are indications that these two circuit configurations

Volume 3 Issue 3 (2024) 2 doi: 10.36922/an.3188