Revise the Backpropagation algorithm from Table 4.2 so that it operates on units using the squashing function tank in place of the sigmoid function. That is, assume the output of a single unit is o = tanhii - 5}. Give the weight update rule for output layer weights and hidden layer weights. Hint: tanth) = 1 tanh2(:.r:). Comments : - This question is similar to Question 2 from H'Wl in that it is asking you to change the neuron's activation function. But for this question, the activation function is tank, and the algorithm is Backpr'opagation (BP), shown in Table 4.2 (p. 93). - (Note: The non-linear BP algorithm is formulated differently from the algorithm for the linear unit in Table 4.1, so your description should be more like the one in Table 4.2, but with differences in how you propagate the error backward.) - Before giving the new algorithrnfsteps, SHOW YOUR DERIVATION of the weight update rule for rm. Then show how it comes into play in the algorithm with 61