Human brain gives the presence of neural network that can accomplish cognitive, perceptual and control tasks. The brain has the ability for computationally demanding the perceptual acts and control activities. The human brain is a massive collection of more than 10 billion interconnected neurons. There is a biochemical reactions to receives, transmit and process information in each neuron.
The cell body in a neuron contains cell nucleus. And the cell body attach with the dendrites. Axon are the extending from the cell body, and are branches into strands and sub-strands. The synaptic terminals of other neurons connected to the axons.
Developed Artificial Neural Network (ANN) as generalization of mathematical model of biological nervous system. The basic elements of neural network is neurons/nodes. Transfer function in the mathematical model of neuron denote the nonlinear characteristic of neurons. And the connection weights that balance the related input signals denote the synapses of neurons. The weighted sum of the input signals indicates the neuron impulse. The learning capability of neuron are achieved by adjusting weights.
Artificial Neural Network (ANN) is a human brain model with machine learning approach and consist of artificial neurons. Neurons in Artificial Neural Network tend to have fewer connection than biological neurons.
In ANN each neuron receives number of inputs and apply activation/transfer function to these inputs which results the output value of the neuron. The architecture of Artificial Neuron Network consists Input layer, Hidden layer and Output layer. The input layer communicates with external environment and inputs a pattern to ANN. As a result, the output layer produce another pattern. The input layer represent the purpose of training the neural network. In artificial neural network the layer which presents a pattern to the external environment are the output layer. Assign the number of output units based on the type of the work that the neural network is to do. The intermediate layer between the input and output layer. It consist of hidden nodes/hidden neurons and apply the activation function on it.
Applications of ANN
There are various types of applications of Artificial Neural Network.
- System identification and control.
- Game-playing and decision making (backgammon, chess, poker).
- Pattern recognition (radar systems, face identification, object recognition and more).
- Sequence recognition (gesture, speech, handwritten text recognition).
- Medical diagnosis.
- Financial applications (e.g. automated trading systems).
- Data mining.
- Visualization and e-mail spam filtering.
Learning Methods of ANN
A neural network has to be configured such that the application of a set of inputs produces the desired set of outputs. Various methods to set the strengths of the connections exist. One way is to set the weights explicitly, using a prior knowledge. Another way is to train the neural network by feeding it teaching patterns and letting it change its weights according to some learning rule. There are three learning rule in neural networks: supervised learning, unsupervised learning, and reinforcement learning.
In Supervised learning a teacher is present during learning process. In this process expected output is already presented to the network. Also every point is used to train the network.
In Unsupervised learning a teacher is absent during the learning process. The desired or expected output is not presented to the network. The system learns of its own by discovering and adapting to the structural features in the input pattern.
In Reinforced learning method a super-visor is present but expected output is not present to the network. Its only indicates that either output is correct or incorrect. In this method a reward is given for correct output and penalty for wrong output.
Among all these methods Supervised and Unsupervised methods are most popular methods for training the network.
Activation function f is performing a mathematical operation on the signal output and chosen depending upon the type of problem to be solved by the network. Some common activation functions are linear activation function, piece-wise linear activation function, sigmoid function and threshold function.
Linear activation function will produce positive number over the range of all real number.
Piece-wise linear function is also named as saturating linear function. This function has binary and bipolar range for saturating limits of the output. Its range lies between -1 to 1.
Sigmoidal function is non-linear curved S-shaped function. This function is the most common type of activation function used to construct the neural network. Sigmoidal function is strictly increasing function by nature.
Threshold function is either binary type or bi-polar type. If the threshold function is binary type its range is in between 0 to 1 and if the threshold function is bi-polar type its range lies between -1 to 1.