In words
What it is, why it matters, and what it is like.
Why am I learning this?
Forward propagation is the part of a neural network that turns raw data into a prediction. You might know how a single artificial neuron works: it takes inputs, multiplies them by numbers called weights, adds a small constant called a bias, and then applies a squashing rule to keep the result within a specific range. Forward propagation is simply repeating this exact same step for every neuron in every layer of the network, moving from the input side to the output side. Once you understand this, you can build a three-layer network yourself, diagnose why your code crashes due to mismatched data shapes, and prepare for the next step: training the network. Forward propagation is also the engine that runs every time you use a trained model—like checking if a loan application should be approved or getting a response from a chatbot—to generate its answer.
The idea, in plain terms
Imagine a production line in a factory. Raw materials enter at one end, and finished goods exit at the other. Each workstation takes what it receives, performs a specific transformation, and passes the result to the next station. In a neural network, the raw materials are your input numbers, such as a loan applicant’s income, debt level, and years of employment. Each layer acts as a workstation. It takes the numbers from the previous layer, multiplies them by its own set of weights, adds its bias, and applies an activation function to produce a new set of numbers. These new numbers become the input for the next layer. This chain continues until the final layer produces the network’s prediction. The key insight is that the transformation at each layer is always the same small operation you already know: multiply, add, and squish. Forward propagation is just performing this operation repeatedly, layer after layer, from start to finish. Nothing more, nothing less.
An analogy
Think of a relay race where each runner does not simply carry the baton but actively transforms it. At the start, you have a number, such as 5. The first runner follows a rule: 'Take the number, multiply it by 0.9, add 0.2, and if the result is positive, keep it; otherwise, change it to zero.' The runner hands this new value to the next runner, who applies a different set of rules. This continues until the final runner produces the race’s result—the prediction. In this analogy, each runner is a neuron in a layer, and their rules are determined by the weights, bias, and activation function. The baton changes value at every hand-off, encoding more meaningful information for the final decision. However, the analogy stops here: in a real relay, runners follow similar paths; in a network, each neuron has unique, learned rules, and the final output is often a probability between 0 and 1 or a set of scores representing different options, not just a race time.
Definition
Forward propagation is the process of calculating a neural network's prediction by passing input data through each layer in sequence, where each layer applies a weighted sum plus a bias to its inputs, followed by a non-linear activation function, with the output of one layer becoming the input for the next.
Where this sits
This concept extends directly from the Artificial Neuron, which you know computes a weighted sum, adds a bias, and applies an activation. Forward propagation is simply performing that calculation across many neurons organized into layers. It also connects to Activation Functions, as you will apply rules like ReLU (which outputs zero for negative numbers) at each stage. The mathematical core involves computing the weighted sums of inputs, which you can think of as combining data points according to learned importance values.
