Skip to content
ml-recall

#update-gate

2 questions from Recurrent Neural Networks

Timed test

2 questions · 2 min · score and explanations at the end

Take a test
Q 1 / 2Score 0 / 0

Question 1: Single choice

hardAI-checkedRecurrent Neural Networks

A GRU computes H~s=tanh⁡(XsWxh+(Rs⊙Hs−1)Whh+bh)\tilde{\mathbf{H}}_s = \tanh(\mathbf{X}_s\mathbf{W}_\textrm{xh} + (\mathbf{R}_s \odot \mathbf{H}_{s-1})\mathbf{W}_\textrm{hh} + \mathbf{b}_\textrm{h}) and Hs=Zs⊙Hs−1+(1−Zs)⊙H~s\mathbf{H}_s = \mathbf{Z}_s \odot \mathbf{H}_{s-1} + (1 - \mathbf{Z}_s) \odot \tilde{\mathbf{H}}_s. You want its output at step tt to depend only on the input Xt′\mathbf{X}_{t'} at an earlier step t′<tt' < t. Which gate values achieve this?

Choose one option