Propagation
Last Updated: July 29, 2026 | By Mihail Sebastian | AI Dictionary
The movement of data through a neural network: forward propagation computes predictions, backpropagation carries error signals back to update weights.
What is Propagation?
Propagation is the movement of information through a neural network. It runs in two directions: forward propagation passes input data through the layers to produce a prediction, and backpropagation passes error signals back through the layers to correct the weights.
The two directions together form the training loop. Forward propagation answers “what does the network predict?”; backpropagation answers “how should it change?”
Types of Propagation
- Forward propagation feeds an input through the network layer by layer. Each neuron combines the previous layer’s outputs using its weights and an activation function, until the output layer emits a prediction.
- Backpropagation starts from the prediction error and works backward, computing how much each weight contributed to it. Gradient descent then uses those gradients to adjust the weights.
Forward Propagation vs Backpropagation
Forward propagation computes the network’s prediction; backpropagation computes how to correct it. The forward pass runs every time the model is used, in training and in production, while the backward pass exists only during training. A deployed model never backpropagates – its weights are frozen.
| Criterion | Forward Propagation | Backpropagation |
|---|---|---|
| Direction | Input layer to output | Output layer back to input |
| When it runs | Every prediction, training or deployed | Training only |
| What it computes | Activations and the final prediction | Gradient of the loss for each weight |
| What follows | A prediction (and, in training, a loss score) | A weight update |
Example of Propagation
A network in training receives a photo of a cat. Forward propagation carries the pixels through the layers, and the output reads: dog, 90% confidence.
The label says cat, so the loss function scores the miss. Backpropagation carries that error backward through the same layers, computing each weight’s share of the blame, and the weights shift to reduce it.
On the next pass, the same photo pushes the output closer to “cat.” Every training step in deep learning is this round trip: forward to predict, backward to learn.
Related AI terms: Backpropagation · Neural Network · Gradient Descent · Weights · Deep Learning
Did you like the Propagation gist?
Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.
Mihail Sebastian — Writes about AI governance, regulation, and the technology behind them. Placeholder bio — replace with a real credential line. About