46838-s99 Machine Learning for Computational Finance
Assignment 4
Due Before Class Monday April 12th, 1999
- Problem 4.1 from the textbook
- (a) Design a two-input perceptron that implements the boolean function ((NOT A) OR B).
- (b) Design a two-layer network of perceptrons that implements (A AND B) OR ((NOT A) AND (NOT B))
- Consider a decision tree with exactly one non-leaf node (i.e. can only ask a single question about a single attribute),
and a single linear perceptron.
- (a)Describe the class of functions the linear perceptron
can represent.
- (b) Describe the class of functions this single-node
decision tree can represent.
- (c)
Can these represent the same functions? Is the decision tree
more powerful than the perceptron? Is the perceptron more
powerful than the decision tree? (i.e. are there any functions
one can represent that the other cannot?) Provide examples that
demonstrate your answer, or a proof.
- In class both perceptrons (linear threshold) and sigmoidal units
were presented.
- (a) How are perceptrons and sigmoidal units similar?
- (b) How are perceptrons and sigmoidal units different?
- (c) Can you learn a non-linearly-separable function with a single
perceptron?
- (d) Can you learn a non-linearly-separable function with a network of
perceptrons?
- Question 4.9 from the text-book.
- In assignment 2 we talked about handling counter-intuitive rules generated by a decision tree learning algorithm.
- (a) How would you recognize an counter-intuitive rule produced
by learning a neural network?
- (b) What could you do about it? (Modify the network? Test the rule? Modify your use of the output?) Explain and justify your decision.
Rosie Jones
Last modified: Mon Apr 5 18:31:45 EDT 1999