BOHR Journal of Biocomputing and Nano Technology

2023, Vol. 1, No. 1, pp. 9–16

DOI: 10.54646/bjbnt.2023.02
www.bohrpub.com

METHODS

Comparative study of single biological neuron with an

artificial neuron
Sashmita Panda *
G. S. Sanyal School of Telecommunications, IIT Kharagpur, Kharagpur, India

*Correspondence:
Sashmita Panda,
[email protected]

Received: 03 February 2023; Accepted: 09 February 2023; Published: 21 February 2023

A number of artificial neural models have been presented in the literature in an effort to suggest a more accurate
representation of a single biological neuron. There are numerous publications on synthetic neurons that attempted
to replicate a single biological neuron, however, such models were unable to generate the spiking patterns of
a real biological neuron. Therefore, there is still scope to design and research improved spiking neural models
that more accurately reflect the functions of a biological neuron. This motivation drives extensive modification of
an artificial neuron model to produce the spike patterns of a real biological neuron. The modified single artificial
neuron model that has been proposed exhibits the functions of a biological neuron. It’s still crucial to model spiking
bio-neuron behavior. Modeling a spiking bio-neuron is still an important exercise in view of possible applications
of the underlying features in the areas of neuromorphic engineering, cognitive radio, and spiking neural networks.
Keywords: artificial neuron, biological neuron, artificial neural network (ANN), back-propagation (BP) algorithm,
spike response model

Introduction action potential or spike reaching a presynaptic neuron’s

axon terminal, allowing Ca2+ ions to enter. With the
A biological neuron is the basic dynamic element of aid of Ca2+ ions, the vesicles are fused to the neuronal
the human central nervous system. When a neuron is membrane, and finally, neurotransmitters are released into
triggered above a certain threshold value, it produces a the synaptic cleft.
brief electrical pulse known as a “spike” (1). The synapse, Numerous studies are currently being conducted on
a fluidic area, is where a spike is transferred from one how our brain’s neurons represent stimuli, and it has
neuron to another (2). In response to chemical and been suggested that the timing of the action potentials or
other inputs, neurons are highly specialized for producing spikes that these neurons release contains information (5).
and sending electrical signals (3). A neuron model that Reaching bio-inspired nanoscale paradigms requires a basic
accurately reflects the biological characteristics of the
understanding of neuro-spike transmission, which is the
Hodgkin-Huxley model and the computational effectiveness
basic mode of communication between neurons. In the paper
of the integrate and fire model is provided in (4). This
(6), the author developed a plausible model to explain neuro-
model reproduces the spiking and bursting behavior of
recognized types of cortical neurons. Neurotransmitters and spike communication. Hodgkin and Huxley explained how
electrochemical impulses help neuro-spike communication. variations in the conductance of Na+ and K+ in the axon
At an axonal terminal of a neuron, synaptic vesicles with membrane can be used to create ionic currents in the giant
neurotransmitters are found. The synaptic cleft is a tiny squid axon (7). They built a mathematical model based
fluid-filled space between an axon terminal of a pre-synaptic on the voltage and time-dependent characteristics of the
neuron and a spine of a different post-synaptic neuron. Na+ and K+ conductance by carrying out several voltage
Voltagegated Ca2+ channels become active in response to clamp tests (8). Through this study, a system of differential

9
10 Panda

aligns was developed, eventually referred to as the Hodgkin-

Huxley model (9), which defined the ionic foundation of the
action potential.
Some cortical neurons spiking and bursting behavior is
described by the model (4). It combines the computational
efficiency of the integrate-and-fire neuron with the dynamics
of the Hodgkin-Huxley model. This model allows for the
real-time simulation of tens of thousands of spiking cortical
neurons. This model is developed in paper (10) using
first-order log domain low pass filters and two translinear
multipliers. The spiking patterns of this chip created neuron
model could be observed by altering the input current as
well as the biased voltages and currents. The Mihalas-Niebur FIGURE 1 | Block diagram of a single biological neuron model.
neuron model is also known as the generalized integrate-and-
fire neuron model (11). Many of the spiking and bursting
characteristics displayed by this model are observed in actual
biological neurons (10, 12). It uses straightforward first-order
differential aligns to explain each of the state variables, in
contrast to other simplified Hodgkin-Huxley neuron models
(13). This neuron model has a number of benefits, including
the ability to be bio-physically interpreted, which enables one
to understand what might occur in biological neurons.
Additionally, it offers a methodical technique to
incorporate numerous additional mechanisms and state
variables. The leaky-integrate and fire (LIF) model’s
counterpart, the spike response model (14), likewise
explains how neurons produce action potentials. While
FIGURE 2 | Block diagram of a single artificial neuron model.
integrate-and-fire versions are based on differential aligns
for the membrane potential, the spike response model relies
primarily on filters. This paper discussed a neuron model the other neurons in the network determines the binary
that accurately simulated the membrane voltage dynamics unit’s value in this model (22). To represent the network of
of a biological neuron cell by including a changeable leaky artificial neurons, researchers have built numerous network
resistor and bias current (15). The author in (16) examines a models. Current applications of artificial neural networks
straightforward model that can faithfully depict the spiking (ANN) span a range of social, industrial, financial, and
behavior of a neuron. While capturing a sizable percentage scientific contexts. Among the frequently used techniques
of the complexity of biophysical models, the spike train in these fields are a functional approximation, filtering,
permits the development of models that are significantly less direct modeling or system identification, inverse modeling
complex than those models. A mathematical description of or channel equalization, control, classification, forecasting,
the evolution of membrane potentials and an adaptation pattern recognition, and optimization. Rumelhart invented
current is provided by a two-dimensional neuron model (14, the conventional back- propagation (BP) technique of
17). It is an evolution of the exponential LIF neuron that multilayer artificial neural network (MANN) as a supervised
imitates the upswing with an exponential function and the learning strategy (23, 24) which is a gradient descent local
downswing with a reset condition in an action potential. optimization technique.
The exponential term causes the voltage to increase quickly
when the membrane potential approaches the threshold
voltage (18).
Some of the parameters used in this model also affect Description of single biological
subthreshold adaptation and spike-triggered adaptation. As neuron model
a result of biological models, numerous artificial models have
been created. The first neural network model (19), which The block diagram of a single biological neuron model
uses directed weighted routes to connect the neurons, was presented in Figure 1 comprises some inputs at the dendritic
described by the author in the study (20). A generalized end. Each of jth (1 ≤ j ≤ J) input of ith sequence, xij
model for giving the threshold in the axon with nonlinear is multiplied with its associated gamma-distributed time-
dynamics that depend on time is the McCullouch-Pitts varying synaptic weights, wij (t) to produce the required
neuron (21). The linear sum of the weighted inputs from output ui (t). In this case, each xij represents either a “0”
10.54646/bjbnt.2023.02 11

TABLE 1 | Numerical values for important parameters used in

simulation study.
Development of a single artificial
neuron model equivalent to
Symbols Description Description Typical Value Values
Symbols biological neuron
tp The time instant when. wij (t) has 1msec A mathematical model of a single artificial neuron which is
maximum amplitude. equivalent to a single biological neuron or a spiking neuron
hp The maximum amplitude of wij (t). 1mv is shown in Figure 2. The model consists of the inputs
which occurs at t = tp
X1...XJ...XNB. The input sequence is taken in terms of “0”s
µ Mean of Gamma distributed 0.5
random. variable and “1”s. Each input is connected with their synaptic weights
σ Variance of Gamma distributed. 0.3 W1...WJ...WNB. Now, the inputs multiplied with synaptic
random variable weights are given to summer or an adder which sums all
Ts Sampling Time. 0.5msec the inputs as a linear combiner. Then, the output of an
adder is applied to an activation function or a squashing
function which limits the permission of the amplitude range
or “1,” and each input sequence consists of a J number of of the output signal to some finite value. The output of the
bits. The strength of the connection between two neurons activation function is represented as ϕ (u).
varies in practice and is represented as synaptic weight. Either
the height of the postsynaptic potential or the slope of the
postsynaptic current denotes the amplitude response and is Steps followed by the single artificial
determined by weights. In most of the single neuron models, neuron model
long-term synaptic plasticity is employed, which assumes
constant synaptic weights (25). In this subsection, a sequence of steps is followed to explain
In Figure 1, ui (t), which denotes the membrane potential. the working principle of a single artificial neuron model
If it crosses some threshold value, then action potentials which is equivalent to a biological neuron.
in terms of spikes are generated. Here ui (t = 0) is the Step 1: Take the input sequence in terms of “0”s
initial membrane potential at the beginning of the ith and “1”s. Here we provide the input sequence as
observation interval. X1...XJ...XNB where NB = 20.
The membrane potential, ui (t) for ith input sequence of Step 2: Load the outputs obtained from the biological
biological neuron model is computed as neuron model of vector size (1000 × 1) and stored them
in an excel sheet.
J
X Step 3: Assign the random value of weights in the artificial
ui (t) = xij wij (t) + ui (t = 0) (1) neuron model, then the input sequence will be multiplied
j=1 with the random weights, and then it will get summed by a
summer using the following alignment.
where 1 ≤ i ≤ I, I is total number of observation intervals,
1 ≤ J ≤ J, J is total number of bits in a sequence, NB
X
xij∈ x1, xj,... xJ, The time-varying weight function can be u= x j wj (4)
mathematically expressed as, j=1

t

1− ttp
 Where “u” is the output of the summing junction and
wij (t) = hp fj e (2) 1 ≤ u ≤ NB.
tp
Step 4: The summing junction output is applied to a
fj is the gamma distributed random variable with mean suitable threshold function where a sigmoid function is used
and variance chosen as 0.5 and 0.3 respectively. The symbol as a threshold.
“tp ” represents the time when wij (t) has attained maximum AH
amplitude hp . In the proposed model, the AMPA receptor y1 = φ(u) = (5)
1 + e−B(u−TH)
is considered which has a typical value of tp = 1 ms and
hp = 1mV respectively. The output “y” of the model can be Where y1 is the output of the single artificial neuron model.
expressed as; “AH” is the maximum amplitude of y1, B is a scalar quantity
and TH is the threshold value after which the output
y = φ(ui (t)) (3) y1 decreases.
Step 5: Now we can give an input of vector size 1000 × 20
ϕ(ui (t)) = 1, when ui (t) > = TH otherwise ϕ (ui (t)) = 0, to the artificial neural model, then in the similar ways
when ui (t) < TH, where “TH” represents a certain output explained in step 4 using Eq. (5) we will find the output of
threshold voltage. the artificial neural model.
12 Panda

FIGURE 3 | Plot of output for a single biological neuron model for FIGURE 5 | Plot of error for artificial neuron model for 100 input
100 input patterns. patterns.

FIGURE 4 | Plot of output for a single artificial neuron model for 100
input patterns. FIGURE 6 | Plot of output for a single biological neuron model for
200 input patterns.

Step 6: To train the artificial neuron model, we will give

the output of the single spiking neuron having vector size
1000 × 1 which is explained in step 2. Step 2: Find the error between the output of a biological
and artificial neuron model. Now by using a learning rule,
this error will be feedback to the artificial neuron model in
Training of artificial neuron model order to update the weights in accordance to get the actual
output which will be nearly equal to the desired biological
An artificial neuron model can be trained by the following
neural model output.
steps;
Step 3: The weights can be updated by using the following
Step 1: After providing the output of a biological neuron,
learning rule;
the model will compute the error function as;
1
e = f (y2 , y1 ) = (y2 − y1 )2 (6) δe
2 1W = −η (7)
δw
Where y2 represents the output of the target or desired
output, here y denotes the output of the single biological
neuron, and y1 is the actual output of the single Where η is the learning rate parameter and e is the error
artificial neuron model. function of the model and its value lies between 0 to 1.
10.54646/bjbnt.2023.02 13

FIGURE 7 | Plot of output for a single artificial neuron model for 200 FIGURE 9 | Plot of error for an artificial neuron model for 500 input
input patterns. patterns.

FIGURE 8 | Plot of output for a single artificial neuron model for 200 FIGURE 10 | Plot of output for a single biological neuron model for
input patterns. 500 input patterns.

Development of updated weights for the Now updating the weights using the learning rule defined in
single artificial neuron model Eq. (7)

δe
In this subsection, the learning rule of updated weights of the 1W = −η
proposed AN model is derived. The output of the summing δw
junction is expressed in Eq. (4); By using the chain rule Eq. (7) can be written as follows;
NB
X δe
u= xJ wj 1W = −η
j=1 δw

Now, the output of the single artificial neuron model is

δe δy1 δu
   
represented in Eq. (5); =η
δy1 δu δwj
AH
y1 = φ(u) = −B(u−TH)
1+e We know;
Error function of the model is given in Eq. (6);
1 δe 1
2
e = f (y2 , y1 ) = (y2 − y1 )2 = y2 − y1 (8)
2 δy1 2
14 Panda

From the model,

δu
= xj (11)
δwj
Hence,
δe
1wj = −η
δwj

δy1
   
δe δu
=η δy1 δu δwj

= η(y1 − y2 )AH B(u − TH)y1 (1 − y1 )xj (12)

FIGURE 11 | Plot of error for an artificial neuron model for 500 input ife = y1 − y2
patterns. andδ = ey1 (1 − y1 )
Finally, the change of weights in the jth branch is derived
as;
= (y2 − y1 ) (9)
1wj = ηδAH B(u − TH)xj (13)

δy1 δ
 
AH Now, using the learning rule,
=
δu δu 1 + e−B(u−TH)
WNew = Wold + 1W (14)

AH B(u − TH)e−B(u−TH) Where WNew is a new weight, Wold is the older weight, and
= 1W is the updated weight for the model. Substituting the
(1 + e−B(u−TH) )2
value of 1wj in the jt h branch the updated weight can be
given as;
1 + eB(u−TH) − 1
= AH B(u − TH) Wj = wj + 1wj
(1 + e−B(u−TH) )2 (15)
Wj = wj + δAH B(u − TH)xj
! Where Wj is a new weight and Wj is the old weight for the
1 + eB(u−TH) − 1 1 artificial neuron model.
= AH B(u − TH) −
(1 + e−B(u−TH) )2 (1 + e−B(u−TH) )2 Step 4: Now weight of the model which is determined in

1 1
 step 3 is applied for finding new weights.
= AH B(u − TH) − Step 5: The training procedure will continue through
(1 + e−B(u−TH) ) (1 + e−B(u−TH) )2
updating the weights till the error function e will be
minimum i.e. the output of both single artificial neuron and
= AH B(u − TH)(y1 − y22 ) biological neuron model will be approximately equal to ∈
(e ≤ ∈ assume ∈ = 0.0001).

= AH B(u − TH)y1 (1 − y1 ) (10)

AH Results and discussions
Where y = φ(u) = 1+e−B(u−TH)
In this section, the results of output between a single
biological neuron (BN) and an artificial neuron (AN) model
TABLE 2 | Comparison of the number of spikes generated in BN and
are analyzed and contributions of the study have been
AN Model with different input patterns.
highlighted. For simulating the proposed models a total of
Number of Input Number of Spikes Number of Spikes MSE 50 input sequences or patterns have been applied to both
Patterns in BN Model in AN Model BN and AN models. By applying a threshold voltage of
100 95 92 0.063
15 mV to the BN model, specific spike patterns at the output
200 38 40 0.066
have been observed. The typical numerical values used for
500 18 18 0.0047
the simulation study of the proposed model are listed in
Table 1.
10.54646/bjbnt.2023.02 15

For experimental-based simulation study, input random References

binary sequence each of length J (J = 20) is considered for
a time interval of 5 ms. Each input bit is then multiplied with 1. Schliebs S, Kasabov N. Evolving spiking neural network—a survey. Evolv
the time-varying synaptic weights obtained at an interval Syst. (2013) 4:87–98.
of TS ms. After each successive TS ms time interval, all 2. Kandel ER, Schwartz JH, Jessell TM, Siegelbaum S, Hudspeth A.
Principles of neural science. 5th ed. New York, NY: McGraw-hill
the twenty partial products are added at the summer node (2013).
to produce an output. After a time interval of 5 ms, the 3. Dayan P, Abbott LF. Theoretical neuroscience: computational
cumulative sum of outputs due to all ten weights are added and mathematical modeling of neural systems. Cambridge, MA:
to produce a membrane potential. When the magnitude Computational Neuroscience Series (2001).
of this potential crosses some predefined threshold value, 4. Izhikevich EM. Simple model of spiking neurons. IEEE Trans Neural
Netw. (2003) 14:1569–72.
then the neuron gets fired, and a spike is produced in
5. Gütig R, Sompolinsky H. The tempotron: a neuron that
the final output. learns spike timing–based decisions. Nat Neurosci. (2006)
It is observed from Figure 3 that after applying 100 input 9:420–8.
patterns to a BN, it generates 95 spikes whereas in the AN 6. Ramezani H, Akan OB. A communication theoretical model- ing of
model it gives 92 spikes shown in Figure 4. The mean axonal propagation in hippocampal pyramidal neurons. IEEE Trans
square error reduces to 0.063 for an artificial neuron model Nanobiosci. (2017) 16:248–56.
7. Abbott L, Kepler TB. Model neurons: from hodgkin-huxley to
presented in Figure 5 which represents an artificial neuron
hopfield. in Statistical mechanics of neural networks. Berlin: Springer
that can mimic a biological neuron. (1990).
Similarly when the number of input patterns increases 8. Koslow S, Subramaniam S. Hodgkin-huxley models. Electrophysiological
to 200, then the spikes generated by the BN model Models In: Databasing the Brain: From Data to Knowledge. London:
is 38 shown in Figure 6 whereas the AN model Wiley (2004).
produces 40 spikes represented in Figure 7. From 9. Hodgkin AL, Huxley AF. Currents carried by sodium and potassium
ions through the membrane of the giant axon of loligo. J Physiol. (1952)
Figure 8, it is seen that the error in the AN model 116:449–72.
reduces to 0.066. 10. Van Schaik A, Jin C, McEwan A, Hamilton TJ, Mihalas S, Niebur E.
Now after increasing the number of input patterns to 500, A log-domain implementation of the mihalas-niebur neuron model.
both the BN model and AN model give 18 spikes which are Proceedings of 2010 IEEE International Symposium on Circuits and
Systems. New York, NY (2010).
shown in Figures 9, 10 respectively.
11. Varghese V, Molin JL, Brandli C, Chen S, Cummings RE. Dynamically
It is observed that the mean square error decreases
reconfigurable silicon array of generalized integrate- and-fire neurons.
to 0.0047 for the AN model shown in Figure 11 and Proceedings of 2015 IEEE Biomedical Circuits and Systems Conference
hence an artificial neuron functions the same as an (BioCAS). New York, NY (2015).
artificial neuron. It can be illustrated that with an increase 12. Jolivet R, Lewis TJ, Gerstner W. Generalized integrate and-fire
in the number of input patterns to both the BN and models of neuronal activity approximate spike trains of a detailed
model to a high degree of accuracy. J Neurophysiol. (2004)
AN model the error decreases and the proposed AN 92:959–76.
model shows the realistic characteristics of a biological 13. Indiveri G, Linares-Barranco B, Hamilton TJ, Van Schaik A,
neuron. Table 2 illustrates that when the number of Etienne-Cummings R, Delbruck T, et al. Neuromorphic silicon
input patterns increases, the artificial neuron exhibits a neuron circuits. Front Neurosci. (2011) 5:73. doi: 10.3389/fnins.2011.
00073
spiking behavior that is quite comparable to that of a true
14. Gerstner W, Kistler WM, Naud R, Paninski L. Neuronal dynamics: From
biological neuron.
single neurons to networks and models of cognition. Cambridge, MA:
Cambridge University Press (2014).
15. Wang Z, Guo L, Adjouadi M. A generalized leaky integrate- and-fire
neuron model with fast implementation method. Int J Neural Syst.
Conclusion (2014) 24:1440004.
16. Teeter C, Iyer R, Menon V, Gouwens N, Feng D, Berg J, et al. Generalized
An improved biological neuron model is proposed and leaky integrate-and-fire models classify multiple neuron types. Nat
its output spike patterns are presented in this paper. Commun. (2018) 9:1–15.
By applying the same input patterns the response of 17. Hertäg L, Hass J, Golovko T, Durstewitz D. An approximation to the
adaptive exponential integrate- and-fire neuron model allows fast and
the biological neuron model is compared with the predictive fitting to physiological data. Front Comput Neurosci. (2012)
proposed AN model. The comparison reveals that the 6:62. doi: 10.3389/fncom.2012.00062
mean square error during the training phase of an 18. Brette R, Gerstner W. Adaptive exponential integrate-and-fire model
artificial neural model decreases as the number of input as an effective description of neuronal activity. J Neurophysiol. (2005)
patterns or sequences increases, implying that an artificial 94:3637–42.
19. Chakraverty S, Sahoo DM, Mahato NR. Mcculloch–pitts neural
neuron model acts more like a biological neuron. The
network model. Concepts of Soft Computing. Berlin: Springer
simulation findings also show that the output of the (2019).
artificial neuron model is almost similar to that of a 20. Mehlig B. Artificial neural networks. arXiv [preprint] (2019) doi: 10.
biological neuron. 48550/arXiv.1901.05639
16 Panda

21. Szu H, Rogers G. Generalized mcculloch-pitts neuron model with 24. Wilamowski BM, Chen Y, Malinowski A. Efficient algorithm for
threshold dynamics. Int Joint Conf Neural Netw. (1992) 3:227119. training neural networks with one hidden layer. IJCNN’99. International
22. Takefuji Y, Lee K. An artificial hysteresis binary neuron: A model Joint Conference on Neural Networks. Proceedings. Piscataway, NJ
suppressing the oscillatory behaviors of neural dynamics. Biol Cybern. (1999).
(1991) 64:353–6. 25. Panda S, Ganguly C, Chakrabarti S. Spiking neuron model with gamma-
23. Rumelhart DE, Hinton GE, Williams RJ. Learning internal distributed synaptic weights for different thresholds. Proceedings of the
representations by error propagation. San Diego, CA: University of 2019 10th International Conference on Information, Intelligence, Systems
California (1985). and Applications (IISA). Piscataway, NJ (2019).