1.slide 19 and 2.slide 20

Single (Hidden) Layer Neural Network zj=w0,j(1)+i=13xiwi,j(1),yk=w0,k(2)+j=1k=1,2,3,4.4g(zj)wj,k(2), Q1: For the perceptron, what are the model parameters to leam based on the training dataset during the training process? | a. Weights | b. Bias c. Nonlinear activation function | d. Output QUESTION 2 Q2: What is the purpose of activation function in neural networks? a. Dimensionality reduction b. Feature engineering c. Introduce non-linearity to approximate arbitrary complex functions d. Gradient computation QUESTION 3 Q3: For the perceptron with bias and three input feature values, how many model parameters does it have to leam during the training process? QUESTION 4 Q4: To build a powerful neural network for leaming nonlinear relationship in the dataset, we can choose linear activation functions in the perceptron. True False QUESTION 5 Q5: For the multi output perceptron as shown in Slide 19 of Lecture 5 with bias, how many model parameters does it have to leam during the training process? QUESTION 6 Q6: For the single hidden layer neural network as shown in Slide 21 of Lecture 5 with bias, how many model parameters does it have to leam during the training process? QUESTION 7 Q7: Training Neural Networks is difficult True False QUESTION 8 Q8: For training a neural network, which of the followings belong to the hyperparameters we need to set beforehand? a. Learning rate b. Epoch number c. Batch size d. Dropout rate QUESTION 9 Q9: Suppose we have a training dataset of 2000 examples. We split it into batches of 500 examples. Then how many training iterations will it take to complete 10 epochs? QUESTION 10 Q10: What is the purpose of dropout in the neural network? a. Avoid underfitting b. Avoid overtitting c. Improve the non-linearity of neural network models d. Increase the model complexity Multi Output Perceptron Because all inputs are densely connected to all outputs, these layers are called Dense layers zj=w0,j+i=13xiwi,j,j=1,2