Homework 8
Writing Assignment
1. Is deeper better in Deep Learning? Can you give some examples/researches/experimental
results to support or oppose to ”deep”? What parameters/designs/structures should be carefully concerned to obtain high performance in Deep Learning?
2. Try to design a feedforward neural network to solve the XOR problem. The feedforward neural
network is required to have two hidden neurons and one output neuron, and uses ReLU as
the activation function.
3. Dropout is a regularization method that approximates training a large number of neural
networks with different architectures in parallel. During training, some neurons are randomly
dropped out. In Figure 1, the neurons that marked with a red cross will be dropped out
during training. For simplication, the bias of neuron is 0 and omitted in Figure 1. Assume
dropout rate is 0.25. This network uses ReLU as the activation function.
a. What are the values of outputs y1, y2 during training?
b. What are the values of outputs y1, y2 during testing?
Figure 1: Feedforward neural network.
1 Due
1. Due is Nov. 22th, 23:59.
2. Submit a PDF on the canvas.
1
