What is Artificial Neural Network (ANN)?
Dr. Robert Hecht-Nielsen, the inventor of neurocomputers, defines a neural network as −“…a computing system made up of a number of simple, highly interconnected processing elements, which process information by their dynamic state response to external inputs.”
Basic Structure of Artificial Neural Network:
Basically, the structure of ANN is totally based on the idea of working human brain. The brain learns from experience.
A neuron is the most basic element of the human brain. It is a specific type of cell. These type of cells called ‘neurons’ provide us with the abilities to help remember, think, and apply previous activities and experiences to our every action. They are about 86 billion and each of these neurons connects itself with another neuron. A neuron can then send the message to another neuron to handle the issue or does not send it forward.
ANN is composed of multiple nodes. These nodes act as neurons in ANN and help to perform like a human brain. The neurons or nodes are connected by a link and each link is associated with weight. Neurons or nodes can interact with each other through the link. Nodes can take input data, process it and the results are sent to other nodes. The output at each node is called its activation or node value.
Types Of Artificial Neural Networks:
1. Feed Forward Artificial Neural Network-
In this type of network, the information flow is unidirectional. It may be Single layer feed forward network or Multiple layers feedforward network (in multiple layer feed forward network there are so many nodes in between the input and output nodes and those nodes are called HIDDEN nodes). A one input unit sends information to another output unit. There are no feedback loops. They have fixed inputs and outputs. They are used in pattern generation/recognition/classification.
2. FeedBack Artificial Neural Network-
In this type of network, the information flow is not unidirectional. Here, feedback loops are allowed. They are used in content addressable memories.
3. Recurrent neural network (RNN)-It is a class of artificial neural network where connections between units form a directed cycle. RNNs can use their internal memory to process arbitrary sequences of inputs.
Working of Artificial Neural Network:
We can see in above diagram, it represents a connection between two neurons or nodes and indicates the pathway for the flow of information. The linkages between nodes are the most crucial finding in an ANN. Each link has weight and the integer number controls the signal between neurons or nodes.
If a network can able to give the ‘desired’ output, then no need of altering or adjusting the weight of the node. If it gives ‘undesired’ output then there is a need to alter the weight according to fulfill the need for ‘desired’ output.
LEARNING TECHNIQUES IN Artificial Neural Network:
- Supervised Learning -Learning is done under the supervision of supervisor or expert. For example, pattern recognition, it involves a teacher or trainer which gives the answer to ANN and then ANN comes up with guesses while recognizing. Then, the network compares its guesses with teacher’s correct answer and makes adjustment accordingly to obtain the desired output by removing errors.
- Unsupervised Learning -There is no supervisor or expert the network can train by itself. When no example data set is known answers then this learning technique is used.
- Reinforcement Learning -In this learning technique, it doesn’t know the exact output. It is built on observations. ANN makes a decision by observing its environment.