Forward propagation is the process by which input data passes through a neural network layer by layer to produce an output. It computes activations and predictions before any training update, such as backpropagation, is applied.