1. Using random search, solve the Cracker Barrel Puzzle. (Due January 31)
2. Monte Carlo: Find the probability of winning the Cracker Barrel Puzzel by (a) choosing a hole at random, and (b) choosing at random the next move from the set of available moves. What is the probability of success? Call this probability p. What is the estimate of the difficulty of the problem in bits, i.e. what is the endogenous information of the problem.
Next, find the probability of winning if you always start with hole #1 empty. Call it q. Then hole #2. Call it q. Etc. Some of the holes will have probabilities greater than p. Others less than p. Compute the exogeneous information and the active information for each in bits. Which hole or holes gives the greatest active information?
Number 3 & 4 on the list above gives required background (Also presented in class). Due Feb 9.
3. Iris data classification. Use the versicolor and virginica iris data. Train an (unlayered) perceptron to classify the two types of irises. Then use the Fisher discriminant to do the same thing. Look at the ROC curves for each. Is there any conclusion you can make? Due Feb 21.
4a. Train the perceptron in Problem 3 using particle swarm.
4b. Show the distributive laws and DeMorgan's laws for min-max fuzzy logic work.