In simple words: A layer is a group of units; hidden layers build intermediate patterns.
Three layer roles
An input layer receives information. Hidden layers transform it through weights and activation functions. An output layer produces a score, category, or number. An Artificial Neural Network (ANN) connects many artificial neurons.
Why use several layers
Early layers can detect small patterns; later layers combine them. A single perceptron cannot solve every pattern. The useful number of layers depends on the problem and the data.
See the flow
input layer→hidden layers→output layer
Simple example
For a handwritten digit, early layers notice short lines. Later layers combine them into the shape of “8”.
Remember: A layer is a group of units; hidden layers build intermediate patterns.