What is Artificial Neural Network

 Artificial Neural Networks are the algorithmic replication of the working of biological neural network.

Artificial neural network is consists of finite number of interconnected nodes, i.e. from few hundred to few thousand, each nodes has their own value, which is called as ‘weight’. Whenever an input is provided to the network the weight is adjusted accordingly to bring the correct output.

The output may not be correct for first few hundred cycles or even thousand, so there would be continuous adjustment of weights through the pre-defined learning rules, until we get correct output.

In artificial neural network each node has input links from other nodes and output links to other nodes, i.e. the output of one node could be input to other, except for input and output units, which has connection with external environment. This type of interconnection helps the neural network to adjust the weights in right direction (i.e. to get correct output).

Artificial neural network; Image source: wikimedia.org

Whenever a node receive input from its neighbor, it does a local computation and then adjusts its own weight, the nodes don’t have global control over the network, however the individual weights of the nodes affect the network’s output.

According to Dr Robert Hecht-Nielsen, the inventor of first neurocomputer, the definition of Artificial Neural Network is:

..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.

Artificial Neural Network is best for the supervised learning. Where you teach a neural network by presenting the sample input and output, when the neural networks learn that, then they can provide the correct output for the input that they never seen.

For example, you can teach a neural network to identify a dog by showing it the pictures of dogs, once they learn, they can identify the dogs, whenever they are presented with the picture of a dog.

The best use of neural network is in Machine Learning. You can teach a computer to do something through neural network that cannot be programmed, or may be programmed but would be a lot more complex to do so.