| Question 1: True or False: Unsupervised learning algorithms require labeled training data? |
| Reference: | Sebastian Raschka and Vahid Mirjalili, Python Machine Learning, Second Edition, Packt Publishing |
| Choice A: | True |
| Choice B: | False |
| Question 2: Among the following, which is not a hyperparameter? |
| Reference: | Sebastian Raschka and Vahid Mirjalili, Python Machine Learning, Second Edition, Packt Publishing |
| Choice A: | Learning rate |
| Choice B: | Number of Layers |
| Choice C: | Weight matrices |
| Choice D: | Size of hidden layers |
| Question 3: Consider a classification algorithm for images. Which type of layer has the fewest learnable parameters? |
| Reference: | Sebastian Raschka and Vahid Mirjalili, Python Machine Learning, Second Edition, Packt Publishing |
| Choice A: | Convolutional layer with ten 3x3 kernels |
| Choice B: | Convolutional layer with three 10x10 kernels |
| Choice C: | Fully connected layer |
| Choice D: | Max pooling layer |
| Question 4: When dealing with a dataset with large class imbalance, which of the following is an appropriate performance metric? |
| Reference: | A Review on Evaluation Metrics for Data Classification Evaluations, Hossin M, Sulaiman MN, International Journal of Data Mining & Knowledge Management Process, Vol. 5, No.2, March 2015 |
| Choice A: | Accuracy |
| Choice B: | Sensitivity |
| Choice C: | Specificity |
| Choice D: | F1 score |
| Question 5: Which of the following is not an advantage of the convolutional neural network (CNN) over the traditional neural network when dealing with image data? |
| Reference: | Sebastian Raschka and Vahid Mirjalili, Python Machine Learning, Second Edition, Packt Publishing |
| Choice A: | Fewer parameters per layer |
| Choice B: | Requires less training data |
| Choice C: | Local-connectivity |
| Choice D: | Parameter sharing |
| Question 6: True or False: Using gradient descent with an appropriate learning rate will guarantee convergence to the global minimum of the cost function. |
| Reference: | An Overview of Gradient Descent Optimization Algorithms, Ruder S, arXiv: 1609.04747v2 [cs.LG] |
| Choice A: | True |
| Choice B: | False |
| Question 7: When convolving a 512-by-512 matrix with a 3-by-3 kernel with no padding, what is the size of the output matrix? |
| Reference: | Sebastian Raschka and Vahid Mirjalili, Python Machine Learning, Second Edition, Packt Publishing |
| Choice A: | 512 |
| Choice B: | 510 |
| Choice C: | 511 |
| Choice D: | 513 |
| Question 8: If you are building a convolutional neural network using a dataset of 100 images, what would be an appropriate distribution for your training and test set if you were to use the holdout method? |
| Reference: | Sebastian Raschka and Vahid Mirjalili, Python Machine Learning, Second Edition, Packt Publishing |
| Choice A: | 95 training images and 5 test images |
| Choice B: | 10 training images and 90 test images |
| Choice C: | 80 training images and 20 test images |
| Choice D: | 50 training images and 50 test images |
| Question 9: In a neural network, if the previous layer has 5 nodes and the next layer has 10 nodes, what is the size of the weight matrix between the two layers? |
| Reference: | Introduction to multi-layer feed-forward neural networks, Svozil D., Kvasnicka V., Pospichal J., Chemometrics and Intelligent Laboratory Systems, November 1997 |
| Choice A: | 10 x 5 |
| Choice B: | 15 x 5 |
| Choice C: | 5 x 1 |
| Choice D: | 10 x 1 |
| Question 10: In a neural network, if the previous layer has 5 nodes and the next layer has 10 nodes, what is the size of the bias vector between the two layers? |
| Reference: | Introduction to multi-layer feed-forward neural networks, Svozil D., Kvasnicka V., Pospichal J., Chemometrics and Intelligent Laboratory Systems, November 1997 |
| Choice A: | 5 x 10 |
| Choice B: | 10 x 5 |
| Choice C: | 5 x 1 |
| Choice D: | 10 x 1 |
