Learning visualised
How does a neural network learn from mistakes?
Show examples. Measure errors. Adjust weights. Again and again.
The animation visualises the training process synchronously across the entire network — including prediction, error calculation and backpropagation.
In detail, it becomes visible how individual connections are adjusted and the system thereby produces gradually better results.
Animation: how a neural network is trained, with clear synapse marking
Step 1 · 0%
training through the network · with detail view
Step 1 — Untrained network
In the beginning, all connections are random. The network knows nothing yet.
Epoche
0 / 100
Fehler (Loss)
—
Vorhersage
—