Comprehensive Introduction to Neural Networks: Advent, Evolution, Applications, and Challenges

Neuron

there are an estimated eighty-six billion neurons in an average human brain. as shown in figure 1, These neurons are themselves connected to each other with a number of structural elements entering them called dendrites, an element leaving them called axons and the central unit called soma, where the information is processed. A typical neuron has a random number of dendrites and axons depending upon the role it plays in a particular function. Figure 1 depicts the basic structure of a neuron (Leterrier, Christophe, 2016).

Figure 1.

­

Neurons channel the incoming signals entering it over the dendrites, carryout computation on those signals at the soma and provoke an output signal on the axon. These input signals are purported to as activations. The axon of one neuron branches out and is connected to the dendrites of many other neurons. The site where a dendrite and a branch of the axon a meet is called a synapse that’s where a neuron transmits signals from one to another. There are estimated to Hundred billion synapses in an average human brain and new synapses are made and broken every second.

Simple Neuron

A key distinctive nature of the synapse is that it can scale the signal (x_i) crossing it. The scaling determinant can be referred to as a weight (w_i). Every synapse is associated with a weight, the signal passing through that synapse gets scaled in response. It is reckoned that the human brain learns is through changes to the weights associated with the synapses. Thus, different responses to the input signals are generated by different weights. the process of learning is fine-tuning the weights in response to a learning provocation, while the network structure does not change. These characteristics of the brain makes it an excellent motivation for a self-Learning-style algorithm. (1) describes a simple weighted sum called Activation function. Where, n = number of inputs, i = i^th input.

In figure 2 An Artificial neuron as depicted in draws its inspiration from the simplified model of biological neuron which was first introduced in the early 1960s. Here, the central node is referred to neuron and the lines directed towards a node are called edges. Edges with weights are analogous to synapses and dendrites. The single line coming out of the node is output which carries the activation. The output is then propagated to other neurons connected to it (K. Rajesh Sai, Plabini Jibanjyoti Nayak and Yallamandaiah S, 2019).

Figure 2.

­