Forward Propagation

Forward propagation (also called the forward pass) is the computational process in a neural network where input data passes sequentially through the network's layers to produce a final output. It is the first step in the training process and a critical part of how neural networks function. In this process, each node in a layer applies weights to the outputs of the previous layer and passes the result through an activation function before sending it to the next layer. The data flows from the input layer → hidden layers → output layer, producing an output suited for the task at hand, such as a prediction or classification. Main steps of forward propagation: • Compute a linear combination of inputs and weights (e.g., z = w₁x₁ + w₂x₂ + ... + b) • Apply an activation function (ReLU, Sigmoid, tanh, etc.) for nonlinear transformation • Repeat this across all layers to generate the final output The error between this output and the ground truth (the loss function value) then drives the next step — backpropagation — which updates the weights. Forward propagation is used in the following contexts: • Computing outputs during model training • Inference and prediction (weights are fixed) • Evaluating neural network architecture designs In short: forward propagation is "creating output from input," while backpropagation is "correcting weights using output error." By repeating these two steps, an AI model adapts to data and becomes capable of accurate predictions.