An activation function is a mathematical function inside a neural network that transforms a layer's input and helps the network represent nonlinear patterns.
Common activation functions influence how a model learns, how gradients flow during training, and whether a network can model complex relationships in text, images, signals, or structured data.