The human inspiration
In our brain, a neuron is activated once it receives in its dendrites an input, a signal strong enough (around 70 mV over less than 2ms) to cause an answer as output.
If activated, the neuron transmits a signal to other neurons through its synaptic terminals
The technological application
Artificial Neurons work “roughly” the same way : they get activated by the information received from the system or from other neurons.
Artificial Neural Networks (ANN) are models that were inspired by the structure and/or functions of biological neural networks.
Nowadays, the parallel is probably less relevant, even if biology remains a souurce of inspiration, as ANN have evolved independently from biological ones and have taken full opportunity of many mathematical developments.
Deep Learning is a modern development of these neural networks allowed by the dramatic increase in computing capabilities. The neural network can include up to thousands of hidden layers, with very complex structures, and each layer hundreds or thousands neurons.
For regression purpose, the output will consist of one single neuron that will output the predicted value.
For classification purpose, there will be as many output neurons as there are classes. Each neuron will output a value that corresponds to the likelihood of its class.
Let’s assume we build a neural network to recognize hand-written digits…
Input (a set of pixels) => Process => Output (as probabilities per class)