Deep Learning Dictionary - Lightweight Crash Course

Universal Approximation Theorem - Deep Learning Dictionary

In general, we can understand that an arbitrary function simply maps inputs to outputs. That's precisely what artificial neural networks do. We supply a network with inputs, and it supplies us with outputs. The network is a function that maps inputs to outputs.

During training, the way in which the network maps inputs to outputs changes as it learns from the data. The point of the training process is for the network to approximate a function that most accurately maps inputs to outputs.