|
Photo Courtesy: Blausen, Wikipedia
|
|
A human brain consists of neurons - cells in the nervous system
of the brain that communicate with each other. A neural network is loosely modeled
to mimic human brain, which consists of millions of interconnected nodes that work
in parallel. In recent years, nodes in neural networks (deep learning) are
organized in multiple parallel layers and allow unidirectional flow of data as
shown in diagram below.
|
|
|
|
A neural network could also be designed to consist of millions of algorithms
(acting like a node) that process data and achieve deep learning.
|
|
|
Types of Neural Network based on their architecture (ASIMOV Institute)
|
|
|
1. Feed-Forward (FF) Neural Network (unidirectional)
|
1.1 Radial Basis Feed-Forward Neural Network (RBF)
|
1.2 Deep Feed-Forward Neural Network (DFF)
|
2. Recurrent Neural Network (RNN)
|
3. Long/Short Term Memory (LSTM)
|
4. Gated Recurrent Unit (GRU)
|
5. Auto Encoder (AE)
|
5.1. Variational Auto Encoder (VAE)
|
5.2. Denoising Auto Encoder (DAE)
|
5.3. Sparce Auto Encoder (SAE)
|
6. Marcov Chain (MC)
|
7. Hopfield Network
|
8. Boltzmann Machine
|
8.1 Restricted Boltzmann Machine (RBM)
|
9. Deep Belief Network (DBN)
|
10. Deep Convolution Network (DCN) or Convolution Neural Network (CNN)
|
10.1 Deconvolutional Network
|
10.2 Deep Convolution Inverse Graphics Network (DCIGN)
|
11. Generative Adversarial Network (GAN)
|
12. Liquid State Machine (LSM)
|
13. Extreme Learning Machine (ELM)
|
14. Echo State Network (ESN)
|
15. Deep Residual Network (DRN)
|
16. Differential Neural Computer (DNC)
|
17. Neural Turing Machine (NTM)
|
18. Capsule Network
|
19. Kohonen Network
|
20. Attention Network
|