Gated Recurrent Unit (GRU)公式简介

update gate $z$: defines how much of the previous memory to keep around.　　

\[z = \sigma (x_t U^z + s_{t-1} W^z )\]

reset gate $r$:determines how to combine the new input with the previous memory.

\[r = \sigma(x_t U^r + s_{t-1} W^r )\]

Cell value $h$: \[h = \tanh (x_t U^h + (s_{t-1} \odot r) W^h)\]

hidden value $s_t$: \[s_t = (1-z)\odot h + z \odot s_{t-1}\]

原文：http://www.cnblogs.com/ZJUT-jiangnan/p/5392923.html