Expanded Explanation
Neural networks are mathematical structures loosely inspired by biological neurons. Layers of nodes apply weighted transformations to an input, and the weights are adjusted during training so the network's outputs match the desired targets.
Simple explanation
Imagine tiny dials that adjust themselves until the network gives the right answer — that adjustment is training.
Technical explanation
A feed-forward neural network is a sequence of linear layers with nonlinear activations. Training uses backpropagation to compute gradients of a loss with respect to weights, updated by an optimiser like Adam.
Simple Examples
- Image classifiers built from convolutional layers.
- Language models built from stacked transformer blocks.