RNN Structure

RNN Structure

The image below shows the structure of a simple RNN.

An RNN maps every $x^{\lt i\gt}$ to a $y^{\lt i\gt}$, and while doing so, it uses information learned at previous timestamps. However, it does not use information from future timestamps.

$a^{<t>} = g(W_{aa}a^{<t-1>}+W_{az}x^{<t>}+b_{a})$

$\hat{y}^{<t>}=g(W_{ya}a^{<t>}+b_{y})$