Backpropagation Through Time
Key concept: Unrolled RNN states flow forward; gradients flow back and add at shared parameters.
Use Next to step through forward, backward, and accumulate. Change sequence length or turn on numbers to inspect the chain.
What to look for: each state combines a local loss gradient with a gated gradient from later steps. The final panel sums contributions to shared θ.
Open the demo in a new tab for more room.