The Backprop Ripple
Watch the Error Travel: A 3D Visualisation of Neural Network Correction.
Backprop Ripple Simulator
The Forward Pass (The Guess)
In a neural network, the Forward Pass is when input data (like an image or a number) flows from the first layer through all hidden layers to the final output. This process results in the network’s final guess.
This tool represents neurons as 3D nodes and weights as glowing connections. Watch the initial signal flow through the network to generate the final prediction.
The Error (The Misjudgment)
After the forward pass, the network compares its guess to the correct answer. The difference is the Error (or Loss). The magic of deep learning happens next: Backpropagation. This tool allows you to click an output node to simulate an error.
You’ll then see the error signal transform into a Ripple that travels backward, mathematically calculating the blame assigned to every single connection and neuron that contributed to the mistake.
The Weight Update (The Learning)
The ripple’s intensity determines how much each connection (weight) needs to change. Large errors cause a strong ripple and a large update; small errors cause a faint ripple and a small update.
The network iterates this process thousands of times, continuously adjusting its weights to reduce the error until its guesses are accurate. This demonstrates the core mechanism by which neural networks learn and adapt.
