Texts: The LSTM Cell
The LSTM cell is able to solve the problem of vanishing gradients because (only one option is correct):
1. The gradient is propagated uninterrupted from the hidden state h to the previous hidden state h(t-1), which ensures that at least some gradient is always propagated backwards in depth.
2. The gradient is propagated from the cell state c(t) to the previous cell state c(t-1) without any weights involved directly between c(t) and c(t-1), which ensures that at least some gradient is always propagated backwards in depth.
3. The gradient is propagated from the cell state c(t) to the previous cell state c(t-1) without any weights involved directly between c(t) and c(t-1), which ensures that at least some gradient is always propagated backwards in time.
4. The gradient is propagated uninterrupted from the hidden state h(t) to the previous hidden state h(t-1), which ensures that at least some gradient is always propagated backwards in time.
Attempt 0 of 2
Submit