back to Introduction to Neural Networks (CSE 40868/60868)

Quizzes

How to solve quizzes?

1. Send your answers to aczajka@nd.edu by the date indicated in each quiz. Please send your answers in plain text and do not use Word or PDF attachments (unless it is necessary).
2. Provide your name or your netID in the email so that I can identify the author.
3. Send your answer quickly. If it is incorrect, we will have some time to develop a correct answer until the deadline.
4. We will discuss shortly the correct answers is class (after the deadline).

Quiz No. 4

Send your answers by Monday, 12/5/2016, 11:59 PM

1. k-means clustering:
   1. is used to train the first layer of the RBF network,
   2. is used to train the second layer of the RBF network,
   3. is not used in training of the RBF network.
2. Consider the same recurrent neural network unfolded to t=10 time steps (case A) and t=20 time steps (case B). The number of parameters to be learned is:
   1. larger in case A,
   2. larger in case B,
   3. the same in both cases.

Quiz No. 3

Send your answers by Friday, 11/4/2016, 11:59 PM

1. Assume that our CNN processes gray scale images (10x10x1). The first convolutional layer has two feature maps and the receptive field is 3x3 pixels. We do not use padding. The output volume of this layer is:
   1. 8x8x1
   2. 10x10x2
   3. 8x8x2
2. Which operation can be interpreted as ensemble learning?
   1. pooling
   2. dropout
   3. zero-padding

Quiz No. 2

Send your answers by Tuesday, 9/27/2016, 11:59 PM

1. We want to use Rosenblatt?s perceptron as a binary classifier for linearly non-separable data. What modification would you apply?
   1. No modifications are required. Rosenblatt?s training algorithm will be very slow in this case, but it will eventually converge to some solution.
   2. Modification to the cost function and making it a margin classifier.
2. Is it possible to construct a single-layer SVM for linearly non-separable data? Justify why (briefly).

Quiz No. 1

Send your answers by Wednesday, 9/7/2016, 11:59 PM

1. We have built a three-layer, fully connected feedforward network (that is: one input layer, two hidden layers and one output layer). Assume that activation functions of all neurons in the first hidden layer are non-linear, and assume that second hidden layer and output layer use affine activation functions. Select a correct answer:
   1. This network is equivalent to two-layer network (that is: one input, one hidden and one output layers).
   2. This network is equivalent to a single-layer network (that is: one input and one output layers).
   3. No simplifications can be made in this network.
2. The learning process in which the network is given the desired responses to all stimuli used in the training is called:
   1. Reinforcement learning
   2. Unsupervised learning
   3. Supervised learning