Home » Soft Computing

Neural Network Architecture in Soft Computing

In this tutorial, we are going to learn about the neural network architecture and also the different classes of neural network architecture.
Submitted by Karishma Das, on May 06, 2020

Artificial Neural Network is defined as a data processing system consisting of a large number of simple highly interconnected processing elements called neurons inspired from the brain. Generally, an Artificial Neural Network structure can be represented using a directed graph.

Diagraph is used in ANN as the neural system is restricted to flow in a specific direction.
A graph is an ordered 2 tuple (V, E) consisting set of vertices and set E of edges.

Classes of network architecture

1. Single layer feedforward network
2. Multi-layer feedforward network
3. Recurrent network

1) Single layer feedforward network

• This type of network comprises two layers, namely the input layer and output layer.
• Input layer neurons receive the input signals and output layer neurons receive the output signal.
• The synaptic links carrying the weight connect every input neurons to output neurons but not vice-versa.
• This type of network is also called feedforward in type or acyclic in nature.
• The output layer alone which performs computations so is also called a single-layer network.
• The input layer sends the signals to the output layer thus the name of the feedforward network.

2) Multi-layer feed forward network

• It comprises multiple layers.
• This type of architecture besides processing an input and output layer.
• This second class of feedforward network distinguishes itself by the presence of one more hidden layer, whose computational nodes are corresponding called hidden neurons or hidden units.
• Hidden layer neurons are present between the input layer and the output layer.
• Hidden layer help in performing useful intermediary computations before directing the input to the output layer.
• Multilayer feed-forward network with L input neurons, m1 neurons in the first hidden layer,m2 neurons in the second layer and n output can be written as: L-m1-m2-n
• These networks differ from feedforward architecture in the sense that there is at least one feedback loop.
• There could also be neurons with self-feedback links, that is the output of neurons is feedback into itself as input

3) Recurrent Network

• These networks differ from feedforward architecture in the sense that there is at least one feedback loop.
• These networks exist one layer with feedback connections.
• There could also be neurons with self-feedback links, that is the output of neurons is feedback into itself as input.