Classification, probability estimation and prediction are important problems across many scientific disciplines. However, the complexity of these problems grows with the domain size of the underlying distributions.

We will study these problems under a competitive framework and provide approaches that work as well as any "natural" algorithm. In particular, we will give results that are independent of the domain size of the problem.

The presentation will cover papers with (subsets of) Hirakendu Das, Ashkan Jafarpour, Alon Orlitsky, Shengjun Pan, and Ananda Theertha Suresh.