Backpropagation algorithm

With the following neural network, we perform backpropagation

using stochastic gradient descent.

Input data are x$=[[1,0]][[0,1]][[1,1]]$ y$=[[0,1,1]].$ All weight values are initialized to $0\mathrm{.}1$ and learning rate is

$0\mathrm{.}2.$

We use sum of squared error $($not mean of squared error$)$ as the

error function $($E$\_($d$)).$ Each node in hidde$($n$)/($o$)$utput has two functions

net$\_($j$)$ and o$\_($j$)($refer to p $\mathrm{.}22$ in the slides$)$ and o$\_($j$)($activation function$)$

is the sigmoid function.

pts