Neural Network Designs

1.Feedforward Neural Network

Neural Network that the cells do not form a cycle.

_images/feedforward.svg

1.1.Multilayer Perceptron

Feedforwrd Neural Network with perceptrons organized into input, hidden and output layers.

1.1.1.Convolutional Neural Network

Multilayer Perceptron with convolutional layers.

_images/convolutional.svg

1.2.Autoencoder

Feedforward Neural Network with equal number of inputs and outputs trained to replicate the input.

_images/autoencoder.svg

2.Recurrent Neural Network

Neural Network that the cells form a directed graph.

_images/recurrent.svg

2.1.LSTM Recurrent Neural Network

Long Short Term Recurrent Neural Networks are Recurrent Neural Networks with LSTM cells with memory.

_images/lstm.svg

2.2.GRU Recurrent Neural Network

Gated Recurrent Unit Neural Networks are Recurrent Neural Networks with GRU cells.

_images/gru.svg

2.3.Neural Turing Machine

Recurrent Neural Network with neural network controller coupled to external memory.

3.Stochastic Neural Networks

Neural Network with random variations introduced to the network via stochastic transfer functions or stochastic weights.

3.1.Restricted Boltzman Machine

Neural Network consisting of input and hidden cells trained to learn the probability distribution of its set of inputs.

3.2.Deep Belief Network

Neural Network consisting of stacking of Restricted Boltzman Machines and their training in a greedy manner.

4.Generative Adverserial Networks

Combination of a generative network and a discriminative network in a competing training.

_images/gan.svg