7CCSMML1 Machine Learning Summer 2017
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
7CCSMML1
Machine Learning
Summer 2017
1.
Name two reproduction operators from genetic algorithms, and explain how they work. (1 mark for each operator named and 4 marks for each explana- tion)
[10 marks]
2.
a. Explain (-greedy action selection.
[5 marks]
b. Consider an agent that is choosing between 3 actions, a1 , a2 and a3 , with the following average rewards: Q (a1 ) = 5, Q (a2 ) = 7, and Q (a3 ) = 4. If the agent uses (-greedy action selection, and ( = 0.1, what is the probability that each action will be selected?
[5 marks]
3.
You have been collecting data on the bars that you have been visiting, and you now have data on nine of them. For each you know what kind of bar it is (Pub, Sports Bar or Wine Bar), the age of the people in the bar (Young, Mixed, Old), how much noise there is (Loud, Normal), whether there is live music (Yes, No), and whether you were happy with going there (Yes, No):
Bar Type Age Noise Music Happy?
B1 Pub Young Loud No No
B2 Pub Young Loud Yes No
B3 Sports Young Loud No Yes
B4 Wine Mixed Loud No Yes
B5 Wine Old Normal No Yes
B6 Wine Old Normal Yes No
B7 Sports Old Normal Yes Yes
B8 Pub Mixed Loud No No
B9 Pub Old Normal No Yes
Explain how you would use a Naïve Bayes classifier to decide whether or not you would be happy going to a new bar, given information on what Type of bar it is, the Age of the people in the bar, the Noise level, and whether there is live Music.
Describe both how you would train the classifier, and how you would classify a new example.
[10 marks]
4.
a. Explain how an agent would use passive reinforcement learning to decide how to act in an unknown environment.
[7 marks]
b. Describe one potential problem with using maximum likelihood to esti- mate the transition model in reinforcement learning.
[3 marks]
5.
a. On the orders of our glorious leader, Emperor Tangerine, you have been building a simple linear classifer to identify whether people are Good Guys or Bad Hombres. The idea is that once it is possible to identify Bad Hombres, they can be deported. You are testing the classifier. Explain how you would use your test results to compute the precision and recall for the classifier.
[6 marks]
b. How would you calculate the F1 score for the classifier?
[2 marks]
c. Emperor Tangerine tells you it is essential that you correctly identify every Bad Hombre, since the security of our country depends on it. What is the likely consequence of doing this?
[2 marks]
6.
a. Write down the algorithm for the K-means clustering algorithm.
[5 marks]
b. What guarantees can be given for the K-means algorithm, and why? [5 marks]
7.
a. How would you would use batch gradient descent to train a univariate linear regression model.
[7 marks]
b. How would the training process differ for a multivariate regression model?
[3 marks]
8.
Explain how Learning from Demonstration (LfD) differs from an Evolutionary Algorithm (EA).
a. Your answer should include information about the representation(s) one might use for each type of method and how training is conducted—i.e., what is learned and when learning takes place.
[6 marks]
b. Your answer should also include an example (or two) that illustrate when it is best to use LfD (2 marks) and when it is best to use EA.
[4 marks]
9.
a. How is a thresholding function used in a neural network, and how does it work?
[6 marks]
b. Give two examples of a thresholding function and explain what each one does. Your explanation could include a drawing of the function’s output.
[4 marks]
10.
a. Write down an algorithm for 10-fold cross-validation.
[5 marks]
b. Explain why 10-fold cross-validation is used in machine learning
[5 marks]
2022-05-23