1. The output of a single neuron is given by y = f(w1x1 +w2x2 +w0x0), where x1 and x2 are inputs, x0 has the constant value 1, w0, w1 and w2 are parameters (weights), and f : R ! R is a function. Suggest a function f typically used in this context. Explain how this neuron could be used to classify inputs (x1; x2) into one of two classes. The input data consists of 8 points:
(1:4; 5) (2:3; 9) (9:6; 9) (0:3; 1) (a0; a1) (a2; a3) (a4; a5) (a6; a7)
where a0 a1 a2 a3 a4 a5 a6 a7 is your student enrolment number. Assign these data points to classes, in such a way that the classification could be learnt by a single neuron. Suggest values for w0, w1 and w2 that would achieve a correct classification, for your choice of function f. Draw a plot of the data points and the boundary line between the classes.
Explain in full detail how the weights of the single neuron in question 1 could be automatically adjusted using gradient descent i.e. how it could be trained to correctly distinguish the classes. Assume f to be the logistic function, and demonstrate why this choice simplifies the numerical calculation of the value of f'.