There is small difference in formulas for GRU in video and slides

Course Q&A Natural Language Processing NLP with Sequence Models
1

Hello, although they would be equivalent in training, i would like to state that the use of Gate-U in the formulas for GRU, is different in the video and slides. The formula makes more sense since the update coefficient is multiplied with the candidata h<t+1>.

image
image953×434 84.6 KB

2

Hi onertan,

Thanks for catching this! In the current version of the video and slides this seems to have been resolved in line with your suggestion.

3

But in the PDF reading, the h^<t_0> and h’^<t_1> are still transposed. The formula and the diagram do not agree.

4

Hi karencfisher,

Thanks for reporting this! I had not checked the reading item. I will report this to the people working on the backend.