Shared weight network in W5A3 - Music inference

Course Q&A Deep Learning Specialization Sequence Models
1

Hi everyone,
Currently, i’m just finished W5A3 about jazz music inference. In the exercise 1, the lstm cell is used same for every x^{<t>} . I wonder if do that then do backward propagation, it would use the current layer to update itself. Does that can cause any problem or mistake in computational process ?

2

Hi @Phan_Phuoc

Hi everyone,

Using the same LSTM cell for every x_t and performing backpropagation through time (BPTT) doesn’t cause any issues. In RNN cells the model to learn temporal dependencies by sharing parameters across all time steps. During BPTT, gradients are computed for each time step and summed to update the shared parameters.

Hope it helps!

1 Like