1. Backpropagation Derive stochastic gradient-descent learning rules for the weights of the net- work shown in Figure 1. All activation functions are of sigmoid form, o(b) = 1/(1+e-6), hidden thresholds are denoted by @j, and those of the output neurons by O;. The energy function is H [t) log M) + (1 - 1) log(1 - 0{)], (1 where log is the natural logarithm, t") are the targets, 0") are the outputs and i labels different inputs. Wjk Wij XK V; 0 Figure 1: Network layout for question 1. 1. Backpropagation Derive stochastic gradient-descent learning rules for the weights of the net- work shown in Figure 1. All activation functions are of sigmoid form, o(b) = 1/(1+e-6), hidden thresholds are denoted by @j, and those of the output neurons by O;. The energy function is H [t) log M) + (1 - 1) log(1 - 0{)], (1 where log is the natural logarithm, t") are the targets, 0") are the outputs and i labels different inputs. Wjk Wij XK V; 0 Figure 1: Network layout for question 1