Browse Teach Curate

Backpropagation Through Time

Training an RNN by unrolling it across time steps into one deep feedforward graph and running standard backpropagation, summing gradients over every copy of the shared weights. In practice the unroll is truncated to a fixed window to bound memory and compute.

jeremy Video

L15.4 Backpropagation Through Time Overview

Backpropagation Through Time (BPTT) is the mechanism that extends the standard backpropagation algorithm to Recurrent Neural Networks by unrolling them across time steps and applying the chain rule t…