What Is Forward Propagation?

How Forward Propagation Works

During forward propagation, information moves in one direction:

The network starts with the input data and continues calculating values until it reaches the output.

Simple Example — Predicting a Student's Result

Imagine we want a neural network to predict whether a student will pass an exam.

We give the network two inputs:

Study Hours = 5
Attendance = 90

These values enter the neural network.

The network processes these values through its layers and eventually produces an output.

Prediction = 0.87

If the output represents the probability of passing, this could be interpreted as approximately an 87% predicted probability of passing, assuming the model is designed and calibrated that way.

Why Is It Called "Forward" Propagation?

It is called forward propagation because information moves forward through the network.

Input
  ↓
Hidden Layer 1
  ↓
Hidden Layer 2
  ↓
Output Layer
  ↓
Prediction

The data does not move backward during this process.

The backward process is called backpropagation, which is used later to calculate gradients and update the model's weights.

What Happens Inside the Network?

A neural network does not simply copy the input to the output. Each neuron performs calculations.

A simplified neuron calculation is:

z = (input × weight) + bias

Then an activation function is applied:

output = activation(z)

You already learned about activation functions in the previous lesson.

A Simple Neuron Calculation

Suppose a neuron receives:

Input = 2
Weight = 0.5
Bias = 1

First calculate the weighted input:

z = (2 × 0.5) + 1

z = 1 + 1

z = 2

Now suppose the neuron uses ReLU.

ReLU(2) = 2

Therefore, this neuron produces:

Output = 2

Forward Propagation Through Multiple Layers

Real neural networks usually contain many neurons and multiple layers.

The output from one layer becomes the input to the next layer.

Input Layer
     ↓
Calculate Hidden Layer
     ↓
Apply Activation Function
     ↓
Hidden Layer Output
     ↓
Calculate Next Layer
     ↓
Apply Activation Function
     ↓
Output Layer
     ↓
Prediction

This process continues until the network reaches the final output layer.

Simple Two-Layer Example

Imagine a very small neural network:

Input
  ↓
Hidden Layer
  ↓
Output Layer

Suppose the input is:

x = 2

The hidden neuron calculates:

z = (2 × 0.5) + 1

z = 2

Apply ReLU:

hidden_output = ReLU(2)

hidden_output = 2

Now this hidden output becomes the input to the output neuron.

output_z = (2 × 0.8) + 0.2

output_z = 1.8

If the output layer uses Sigmoid:

Sigmoid(1.8) ≈ 0.858

So the final prediction is approximately:

Prediction ≈ 0.858

Complete Forward Pass

This entire process is what we call a forward pass.

When the network performs this process for an input, we say that it is performing forward propagation.

Forward Propagation vs Training

Do not confuse forward propagation with the entire training process.

Forward propagation produces the model's prediction.

Input
  ↓
Forward Propagation
  ↓
Prediction
  ↓
Loss

During training, the loss is then used as part of the process that calculates gradients and updates weights.

Prediction
    ↓
Loss
    ↓
Backpropagation
    ↓
Gradients
    ↓
Update Weights

You will learn the backward process in the Backpropagation lesson.

Example — Cat or Dog?

Suppose we build a neural network that classifies an image as either a cat or a dog.

The image pixels enter the network:

Image
  ↓
Input Layer
  ↓
Hidden Layers
  ↓
Output Layer
  ↓
Prediction

Suppose the output is:

Cat = 0.91
Dog = 0.09

The model would choose the class with the higher predicted probability:

Prediction = Cat

The calculation that produced those values is part of forward propagation.

The Big Picture

Input Data
    ↓
Input Layer
    ↓
Weighted Calculation
    ↓
Bias
    ↓
Activation Function
    ↓
Hidden Layer
    ↓
Weighted Calculation
    ↓
Bias
    ↓
Activation Function
    ↓
Output Layer
    ↓
Prediction

The key idea is that the information moves from the beginning of the network to the end.