Artificial Intelligence COSC 6368
Popular in Course
Popular in Chemistry
This 6 page Class Notes was uploaded by Lowell Harris on Saturday September 19, 2015. The Class Notes belongs to COSC 6368 at University of Houston taught by Christoph Eick in Fall. Since its upload, it has received 75 views. For similar materials see /class/208165/cosc-6368-university-of-houston in Chemistry at University of Houston.
Reviews for Artificial Intelligence
Report this Material
What is Karma?
Karma is the currency of StudySoup.
Date Created: 09/19/15
Artificial Intelligence COSC 6368 Qualifying Exam Part Two Thursday December 19 2002 1030a Name SSN l Neural Networks 7 2 Directions in Machine Learning 17 3 Reinforcement Learning 1 l 4 Decision Trees 7 Point Total out of 42 The exam is open books and notes and you have 75 minutes to complete the exam This part will count approx 34 for your qualifying exam grade l Neural Networks 7 a Assume we have to learn a function XfABC with an neural network whose architecture is depicted below Assume the perceptron learning algorithm is used and the current weights for w0 is 02 wl is 03 w2 is 04 and W3 is 05 and the example ll00 is processed this means A and B are both 1 and C and X are both 0 by the learning algorithm How would the algorithm update the weights w0 wl w2 W3 assuming a learning rate at of 05 is used 4 w0new wl new w2new w4new Problem 1 continued b Why is learning multilayer networks more complicated than learning single layer neural networks 3 2 Directions in Machine Learning Dietterich Paper Review a Why are ensembles bene ciary for constructing more accurate classi ers Why does the author of the paper believe that generating ensembles works Limit your answer to 610 sentences 4 b One method for learning ensembles is injecting randomness into the used learning algorithm How could this approach be used to create ensembles of decision trees 3 c D V Problem 2 continued Assume you have to apply inductive learning e g try to learn feed forward neural networks to data sets with more than 10000 attributes features Give a sketch a discussion of 1 0r 2 methods is suf cient what can be done to speed up learning classifiers for data sets that involve a very large number of attributes Limit your answer to 46 sentences 3 On page 24 0f the paper the author mentions that Hence online learning algorithms must balance exploitation with exploration Explain 3 How can belief network technology bene t from work centering on learning stochastic models as discussed in section 5 0f the paper Limit your answer to 46 sentences 4 3 Reinforcement Learning 1 l a Assume the following state space is given in which an agent can execute actions a and b The agent receives a speci c reward for reaching a particular state 9 for reaching state 0 8 for state 1 and 712 for state 2 For example if the agent applies action a in state 0 he reaches state 1 and obtains a reward of 8 Assume Qlearning see page 613 of the textbook is applied to this problem The agent executes the action sequence aaabab how does the Qtable look like after the agent executed the 6 actions assume that the initial Qvalues are 0 and that the learning rate at is 05 Indicate every update in Qvalues below 6 Qa0 o Qb0 o Qa1 o Qb1 o Qa2 o Qb2 o Interpret the QTable you obtained What does the Qtable tell the agent concerning what actions are the most useful 2 b What is are the main differences between inductive learning and reinforcement learning 3 4 Decision Trees 7 We would like to predict the gender of a person based on two attributes legcover has two possible values pants or skirts and weight has 3 possible values heavy medium skinny We assume we have a data set of 20000 individuals 15000 of which are male and 5000 of which are female Skirts are present on 40 of the females and no male wears a skirt 50 of the males are heavy and 50 are medium weight 80 of the females are skinny and 20 are medium weight i Compute the information gain of using the attribute legcover for predicting gender 3 ii What is the information gain of using the attribute weight to predict gender 4