MLP backward
\[\begin{aligned}
&For \space an \space output \space layer \space node \space k\in K\\
&\frac{\delta E}{\delta W_{jk}}=O_j\delta_k\\
&where\\
&\delta_k = O_k(1-O_k)(O_k-t_k)\\
&For \space a \space hidden \space layer \space node \space j\in J\\
&\frac{\delta E}{\delta W_{ij}}=O_i\delta_j\\
&where\\
&\delta_j=O_j(1-O_j)\sum_{k\in K}\delta_k W_{jk}\\
\end{aligned}
\]
