- Introduction to neural networks
- Supervised and Unsupervised Learning
- Feed-forward Backpropagation Neural Networks
- Backpropagation Algorithm Overview
- Learning Rate
- Momentum and Jitter
- Activation Functions
- Kohonen Self-Organizing Maps
- Neighborhood Functions
- Lattice Topology
- Kohonen Layer Shape
- Neural Network Design
- Applications of neural networks
- Setting up the development environment
- NeuronDotNet Class Overview
- Some useful code snippets
- NeuronDotNet Sample Applications
Introduction to neural networks
Artificial neural networks attempt to create a mathematical model of human brain. Human brain is composed of billions of massively interconnected nerve cells called neurons, working in parallel. Information is transferred from one neuron to another in the form of electrical impulses through connections called synapses. A typical neuron is connected to several tens of thousands of other neurons via synapses.
The neural structure keeps changing throughout the human life. A basic structure which is defined at birth, is developed through learning as new connections are made and destroyed. Later changes mainly consist of strengthening and weakening of synaptic junctions. (For example, the process of learning a new friend's face involves changes in strengths of synapses). The large number of neurons, high degree of interconnectedness and complicated behaviour of each neuron is what makes the brain incredible.
Artificial neural networks are extremely simple models of human brain. Artificial Neurons which form the basic building block of an artificial neural network are simple structures that are designed to imitate biological neurons. These neurons communicate via directed artificial synapses.
A typical artificial neural network consists of hundreds of artificial neurons connected together by thousands of artifical synapses. To simplify the whole structure, neurons having similar properties are grouped into layers. No two neurons belonging to the same layer are connected to each other.
The most commonly used three-layered feed-forward neural network architecture is shown above. Artifical neurons with similar properties are grouped into Layers. The first layer is the input layer which gets input from the external environment. For example, in case of face recognition, the bitmap image of the face is fed to the input layer. Each neuron in this input layer is usually connected to all neurons in the next layer (which is called hidden layer). Output layer gets signal from hidden layer and provides the result to the external environment.
The mechanism of data transfer and values of various parameters involved in propagation are suitably adjusted to make the neural network behave like an artificial brain. The process of adjusting these parameters and thereby make the neural network learn, is called Neural Network Training. Various algorithms are available for training a neural network.