A Comprehensive Study on Architecture of Neural Networks and Its Prospects in Cognitive Computing

2. Artificial Neuron: An Abstract Representation

The human brain can be considered as a highly complex structure that can be viewed as a highly connected network of neurons (Neural Network, n.d.; Sivanandam & Deepa, 2011). Accordingly, the biological neuron can be modeled into artificial neuron. Each constituent of the model bears analogy to actual components of biological neuron. Figure 1 shows a simple model of artificial neuron on the basis of which the artificial neural network is built. In the diagram, x₁, x₂,…, x_n represent the n number of inputs supplied to the artificial neurons and w₁, w₂, …, w_n represent the weights concerned with the inputs respectively. Similar to the biological neurons, the whole input received by the artificial neuron I can be denoted as shown in following equation:

Now the above sum gets passed through the non-linear filter Φ known as Activation function or squash function.

In this context, an indubitable activation function is employed called as the threshold function. Here the sum gets compared with the threshold value ɵ. If the value of I is higher than ɵ, them the output becomes 1; otherwise, this becomes 0.