Minimal gated unit variations (`MGU1`, `MGU2`, `MGU3`)

[MartinuzziFrancesco]

Paper

https://doi.org/10.1109/MWSCAS.2017.8053242

Equations

change the gate equations with the following

$$\begin{equation} f_t^j = \sigma \left( (U_f h_{t-1} + b_f)^j \right) \end{equation}$$ $$\begin{equation} f_t^j = \sigma \left( (U_f h_{t-1})^j \right) \end{equation}$$ $$\begin{equation} f_t^j = \sigma \left( (b_f)^j \right) \end{equation}$$

Official implementations

None