Hidden Layer

A neural network typically consists of three kinds of layers: an input layer (takes raw data), one or more hidden layers (process the data internally), and an output layer (delivers the result). The hidden layers are the actual heart: this is where the model learns which features are relevant for the task and how they are combined.

The number of hidden layers determines model capacity. Classic neural networks (before 2010) usually had only one or two hidden layers. Modern deep learning models like GPT have dozens to hundreds of hidden layers, hierarchically building increasingly abstract representations. Early layers recognize simple patterns, later layers complex concepts.

For Pimcore users, hidden layer is a technical concept behind the platform's AI functions. Image tagging, classification, and text generation use deep learning architectures with many hidden layers. Users do not need to understand the internal layers, but they benefit from the model performance these enable.