Types of Neural Network

TYPES OF  NEURAL NETWORKS

The working of neural networks is similar to human brains in terms of decision-making. Neural networks differ from one another due to their unique principles used in framing rules. There are different types of neural networks - Feed-Forward Neural Network, Radial Basis Function Neural Network, Multilayer-Perceptron, Convolution Neural Network, Recurrent Neural Network, etc. The neurons receive input from previous layers, then process it and send it to the upcoming layers. Weights, activation functions plays important role in determining the strength of the input. 

Let us understand how neural networks work. Neural networks working is similar to the nervous system of the human body. It comprises one input layer, two or more hidden layers, and finally one output layer.

Neural Network

• Let us consider an image called “9”, as an input, which is of 26x26 pixels.

• All together 26x26=676 are sent into the input layer. All neurons have a value called an activation function so whenever that number is met that particular neuron will get fired and the value will be sent to the upcoming hidden layers.

• This process is done in all hidden layers. once the hidden layer job is done, control goes to the final layer, this process of flow of control from the initial input layer to the final output layer is known as forward- pass.

• After results are obtained, it will be checked with input image 9 to check whether it matches the variation between the predicted and the actual output. This is deviation is known as the loss function.

• Once the loss value is calculated, the same information is sent back from the output layer to the input layer through those hidden layers. This process is termed as Back-Pass. And, inputs are slightly adjusted and trained the network. 

Types Of Neural Networks

  There are different types of neural networks. Let us understand some of them here.

1. Feed-forward Neural Network

    

                                                                             source

• This is the very basic neural network where the flow starts from the input layer and goes to the output layer.
• They comprise  only one hidden layer.
• Here, the data moves only in 1 direction there is no back-propagation technique in this network.

2 . Multi-layer Perceptron

 

 

• This is network is a supplement of a feed-forward neural network. It is composed of three different layers like Input layer, hidden layers, and output layer.
• In this network, all nodes are fully connected. It is used to classify the non-linear data.
• In the forward propagation, the signal flow moves from the input layer through the hidden layers to the final layer, and the output is compared with the actual values.
• In the backpropagation, using calculus, partial derivatives of the error function w.r.t. the various weights and biases are back-propagated through the MLP.

3. Convolution Neural Network (CNN)

CNN

• CNN is one of the variations of a multi-layer perceptron. It is widely used for object detection and classification.

• It can contain more than one convolution layer and since it contains a convolution layer the network is very deep with fewer parameters.

4. Recurrent Neural Network

RNN

• RNN is a type of network in which the output of one neuron is fed back as input to the same node.

• RNN deals with sequential data, it takes current data and also previous inputs. With the help of Internal memory, RNN memorize the previous input data.

• It is widely used in Image captioning, Time series prediction, NLP, etc.

5. Modular neural network

MNN

• A combination of multiple smaller neural networks defines MNN.

• All the sub-networks make a big neural network and all of them work independently to achieve a common target.
• It is used in Stock market forecasting, high-level input data compression.

Conclusion
                   

In this article, we have seen how neural network functions and also different types of neural networks. We have also seen the intricacies of the working of all those networks and the application of those networks.