The forward pass
Data enters the input layer and flows through each hidden layer until it reaches the output. Each layer multiplies by weights, adds bias, and applies activation.
Input (3 features)
⟶
⟶
Output (sigmoid)
Click Animate to watch values flow through the network.
Before vs after training
Before training: Weights are random → output is random → predictions are meaningless.
After training: Weights are tuned → output matches labels → predictions are useful.
Loading...
Loading...
Loading...
Think Deeper
Try this:
Click 'New random weights' several times. Does the same input always produce the same output?
No — different random weights produce different outputs. Before training, the network is random guessing. Training adjusts weights so the output matches the correct labels.
Cybersecurity tie-in: The forward pass is what happens during inference —
when the model classifies a new connection as attack or benign. It runs in milliseconds,
making neural networks fast enough for real-time intrusion detection.