A perceptron is like a neural network that has no hidden layers and a single output unit with on e of the following activation functions. g(z)=1+ez1g(z)=max(0,z)g(z)={1ifz>01ifz0g(z)=tanh(z) Suppose we have data with four features about movies that we would like to classify into one of three classes: romantic comedy, horror, documentary. Q1: Which of the following two neural networkarchitectures is more powerful (where "more powerful" means it can do everything the other can do plus more)? You need to explain why you choose this network architecture. (5 marks) - A single neural network with 4 inputs, 15 units in a hidden layer, and 3 units in the output layer - Three separate neural networks, each with 4 inputs, 5 units in a hidden layer, and 1 unit in the output layer Q2: Suppose you have a neural network where, for one of the hidden layers, all units in the layer haveidentical values for their weight parameters as the other units in the layer. Which of the followingis true? You need to explain why. ( 5 marks) - When new training data are provided to the neural network and training progresses, the units in the layer will continue to all have identical values for their weight parameters - removing all units except one from the layer would result in an equivalent classifier - all units in the layer will have the same input and output as the other units in the layer - since the neural network has one or more hidden layers, it is possible that it may be able to model any continuous function, e.g., log(x),1/x,sin(x), ex. - all of the above Q3: Consider the neural network below. Ignoring any weight parameters associated with bias terms, how many values for weight parameters would need to be leamed for this neural network? Kindly explain your answer (5 marks) Q4: Consider the neural network below, Including weight parameters associated with bias terms, howmany values for weight parameters would need to be learned for this neural network? Kindly explain your answer ( 5 marks)