Neural networks
A layer is one row of neurons in a neural network; data passes through layers in order, with each layer turning its input into a slightly more useful form.
Neurons in a network are grouped into layers. The input layer receives the raw data, such as pixel values. Hidden layers in the middle do the real processing. The output layer gives the answer, such as a probability for each category. Signals flow from one layer to the next, never skipping ahead in the simplest designs.
Each layer transforms what it receives. In image recognition, the first layer might detect edges, the next combines edges into textures, and later ones assemble textures into parts and objects. None of this is programmed; it emerges from training. 'Deep' learning just means many layers.
Different layer types exist for different jobs. Convolutional layers scan images. Attention layers, used in transformers, let every word in a sentence look at every other word. Modern large language models have dozens of attention layers stacked in sequence.
A face recognition network starts with a layer that sees lines, passes to layers that see eyes and noses, and ends with a layer that says whose face it is.