# Independently Recurrent Neural Network(IndRNN):Building A Longer and Deeper RNN

### The gradient backpropagation throgh time for an IndRNN

$\frac{\mathrm{\partial }{J}_{n}}{\mathrm{\partial }{h}_{n,t}}=\frac{\mathrm{\partial }{J}_{n}}{\mathrm{\partial }{h}_{n,T}}\frac{\mathrm{\partial }{J}_{n,T}}{\mathrm{\partial }{h}_{n,t}}=\frac{\mathrm{\partial }{J}_{n}}{\mathrm{\partial }{h}_{n,T}}\prod _{k=t}^{T-1}\frac{\sigma {h}_{n,k+1}}{\sigma {h}_{n,k}}$$\frac {\partial J_{n}} {\partial h_{n,t}} = \frac {\partial J_{n}} {\partial h_{n,T}} \frac {\partial J_{n,T}} {\partial h_{n,t}}=\frac {\partial J_{n}} {\partial h_{n,T}}\prod\limits^{T-1}_{k=t}\frac {\sigma h_{n,k+1}} {\sigma h_{n,k}}$
$=\frac{\mathrm{\partial }{J}_{n}}{\mathrm{\partial }{h}_{n,T}}\prod _{k=t}^{T-1}{\sigma }_{n,k+1}^{\prime }{U}_{n}$$=\frac {\partial J_{n}} {\partial h_{n,T}}\prod\limits^{T-1}_{k=t}\sigma' _{n,k+1} U_n$
$=\frac{\mathrm{\partial }{J}_{n}}{\mathrm{\partial }{h}_{n,T}}{U}_{n}^{T-t}\prod _{k=t}^{T-1}{\sigma }_{n,k+1}^{\prime }$$=\frac {\partial J_{n}} {\partial h_{n,T}}U_n^{T-t}\prod\limits^{T-1}_{k=t}\sigma' _{n,k+1}$