Give a decision tree representing a simple boolean function

What is the entropy of the labels in a set of data and
what is the information gain of an attribute.

Design a perceptron (or small network of perceptrons)
implementing a simple function.

Derive a gradient descent learning rule for updating 
the parameters in a prediction rule with a given loss measure 
(e.g. minimize the square loss of a function like 
w1*x1 + w2*x2 + w3*(x1)^2 + w4*(x2)^2 on a set of data).

Given a concept class $\cal C$, why might it be easier
to PAC learn $\cal C$ using a 
larger hypothesis class ${\cal C'}\supset {\cal C}$?

According to the (noise-free) PAC sample size bounds, if we want to increase the accuracy by a factor of two, approximately how many more examples will we need?

What is the VC dimension of hyper-rectangles in d dimensions?
(i.e. each hyper-rectangle is described by range for each dimension and a point is in a hyper-rectangle if each coordinate of the point is in the corresponding range).

Given a set of coin flips and a prior distribution on the probability of 
heads, compute the maximum likelihood, maximum a'posteriori, and mean a'posteriori probabilities that the next flip is heads.

Given losses and probabilities of outcomes, find a Bayes optimal prediction.

Find the maximum likelihood gaussian distribution from a (small) sample.

Given some data, find the prediction made by Naive Bayes on a (given) new instance.

Describe the difference between regression and classification problems.

Given a (simple) coding scheme, find the hypothesis which minimizes the description length of the data.

Given a Bayes network over some features, and some relationships between features (indepence, conditional independence) does the network correctly reflect the relationships?

Simulate the WM algorithm on a sequence of instances.

How do the mistake bounds of Winnow for learning disjunctions depend on the number of attributes?

Run the perceptron algorithm on a sequence of instances

What is the logistic function?