Page 118 - AN-3-3

Advanced Neurology Evaluating plausibility of thalamic model

aim here is to subject the model to initial experimental communication with Rs and the extraction of PCs. On
conditions to observe general responses analogous to the other hand, the burst mode, which only occurs with
biological phenomena such as wave sculpting, inhibitory the next depolarization after sustained inhibition for >100
facilitation, hallucinatory experiences, and pattern ms by Rs, enables communication with cortical neurons
completion. This model also extends Llinás et al. proposals operating exclusively at high frequencies. 14
by providing a computational explanation of these Throughout the network training phase, we employed
phenomena, potentially illuminating neurophysiological the incremental version of the probabilistic pre-synaptic
aspects, as well as cognitive and behavioral phenomena. Hebbian rule, as developed by Peláez and Andina. Although
55
Consequently, subjecting the model to experimental the pre-synaptic rule is more suitable for computational
conditions based on empirical findings offers an excellent processes, when attempting to model the fundamentals
approach for both its theoretical and experimental of a probabilistic algebra occurring in the thalamus, we
validation. employed the pure probabilistic interpretation, as shown in

2. Methods the Supplementary File. This version allows the emulation
of the biological characteristics pertaining to synaptic
2.1. Programming environment directionality and the metaplasticity of LTP and depression.
The programming, development, and evaluation of the Put simply, the pre-synaptic rule can be expressed as
modeling were conducted using MatLab software (version (Equation I):
®
R2020b, MathWorks Inc.), which provides a customizable ΔW = ξI (O–W) (I)
environment and flexible control over the network. All
simulations were conducted on computers that met the Where the difference between the probability of a post-
®
recommended requirements for MatLab software. synaptic potential O and the synaptic weight is multiplied
by the learning factor ξ and the probability of the pre-
2.2. The thalamic model architecture synaptic potential I. This synaptic weight modification
rule is grounded in biology, deriving its basis empirically
This is a phenomenological model where the network from the plasticity curve that depicts the correlation
architecture emphasizes the computational aspects of the between post-synaptic voltage and synaptic weight
biological thalamic circuitry by incorporating bioinspired modification. The increment in synaptic weight undergoes
56
characteristics into its design. Given the pluripotent a counterbalancing effect of intrinsic plasticity, which is the
computational capacity of the thalamus, the type of relay dynamic adjustment of the shift in the sigmoidal activation
considered is expected to have minimal impact on the function. The synaptic weight experiences a greater shift as
central set of computational processes in this circuitry. it grows in alignment with the network input value. This
Therefore, we modeled the system using the first-order dynamic adjustment mechanism prevents synaptic weights
relay for this work. from perpetual growth, stabilizing them at specific values.
Fundamentally, the thalamus exhibits a comparable To simulate activity in two modes, we adopted a
architecture to that of an auto-associative neural bioinspired programming philosophy. Figure 6 illustrates
network, as illustrated in Figure 3C. It comprises two the algorithm’s flowchart, delineating the initialization,
layers (Figure 5) featuring excitatory feed-forward REs iteration loops, and the conditions for process termination.
responsible for receiving input patterns and inhibitory During the initial stage, the network parameters are
feedback Rs projecting from the second to the first layers. initialized (Figure 6A): nt = number of iterations,
In contrast to presenting the same pattern to both input determining the total number of iterations; np = number of
and output layers during training, the thalamus employs patterns, specifying the total number of patterns; steps = 40,
a singular input/output layer. The comparison stage of indicating the total number of steps in an oscillatory cycle
the auto-associative network is seamlessly integrated into (arbitrary set at 40); t = 0, initializing iterations at zero; p =
the input/output layer of the thalamus. Comprehensive 0, patterns are denoted as I , where p is an integer greater
p
details of the model have been more extensively elaborated than 0; and threshold = 0.97, signifying the threshold at
elsewhere. 8,11,12 which the first Rs, on reaching activation, becomes the

2.3. Learning rules and model dynamics winner in a winner-take-all process. Next, the weights and
i
shifts are initialized (Figure 6B): R and shift(i) for REs R,
i
All patterns are represented as a 9 × 9 input matrix  j j
equivalent to the number of REs in the network (REs in and O and shift(j) for Rs O .
Figure 5). Biologically, these neurons exhibit two distinct While the current iteration is < the total number
firing modes: tonic and burst. The tonic mode supports of iterations (t < nt), the following oscillation update

Volume 3 Issue 3 (2024) 6 doi: 10.36922/an.3188