We study the usage of information-theoretic measures in learning problems. The first problem considered is the algorithm-dependent generalization error bound. Conceptually, the mutual information between the output of ...