Neural Network and Its Types

Feed-forward Neural Networks.

These are the simplest and the most common type of deep neural network.
They also have many applications.
They were the first type of deep neural network to be used, and they have the typical structure.
There are also a number of non-linear activation functions that are applied.
In these networks, information flows in one direction only. That is, forward. There are no cycles or loops.
You can think of feed forward neural networks as being the basic type of deep neural network.

The flow of information through a recurrent neural network, does form cycles or loops.
Recurrent neural networks have directed cycles where following the flow of information, may result in arriving back at your starting point.
Recurrent neural networks can be very difficult to train due to the complicated dynamics that result from the cycles.
There are many different types of recurrent neural networks. In this we can see that the output can also be used as input to the neural network.
It is often the case that the directed cycles will occur within the neural network.

A convolutional neural network is a type of neural network, which is most often used to perform visual learning tasks.
They're used widely in image and video recognition, in image classification, as well as in natural language processing.
Convolutional neural networks consist of an input layer, an output layer and a varying number of hidden layers.
The hidden layers in a convolutional neural network contain a number of convolutional layers.
We often have a pairing of a convolutional layer with something which is known as a pooling layer.
Convolutional layers summarize features in an input image.
Convolutional neural networks apply filters or kernels that are smaller than the image being processed. These filters are systematically applied over the entire image. This allows features to be identified anywhere on an image.

GANs are deep neural networks that are composed of 2 separate networks.
These 2 networks have an adversarial relationship.
GANs are of great interest at the moment because they are able to be creative.
They can create images, music, speech, and poetry.
GANs can be used to create deep fakes.
The generator network, or the one network will generate images that are used to try and fool the discriminator network.
The discriminator network, which has been trained on real images, will look at the generated images and try and predict whether they are generated,
or they are real.
The discriminator network will predict a label, real or fake.
The output of the discriminator network is in a feedback loop with the generator network.
This allows the generator network to know how well it is doing