Crowdly
Add to Chrome
A simple RNN has 3 time steps, and the gradient of the loss with respect to the ...
✅ The verified answer to this question is available below. Our community-reviewed solutions help you understand the material better.
A simple RNN has 3 time steps, and the gradient of the loss with respect to the hiddenstate at time step 3 is 0.8. If the recurrent weight matrix has an eigenvalue of 0.5,what will be the approximate gradient backpropagated to time step 1 (ignoring inputgradients)?
❌
❌
❌
✅