Classification is one of the most common tasks in machine learning. Within classification, there is normally a split between binary classification (only two possible outputs) and multiclass classification (more than two possible outputs). The theoretical analysis of these settings shares the same split. Under the P AC-learning model, binary classification learnability under the 0-1 loss is known to be characterized by the VC-dimension [V apnik and Chervonenkis, 1974, Shalev-Shwartz and Ben-David, 2014]. For multiclass classification, there has also been a further split between finite and infinite label cases.

In this paper, we study the role of machine-learned predictions in offline NP-hard problems.

Suppose we are considering a multivariate event sequence. What caused a specific event to happen? Which variables are causes of each other?

Learn-to-Defer is a paradigm that enables learning algorithms to work not in isolation but as a team with human experts.

The study of online algorithms with machine-learned predictions has gained considerable prominence in recent years.

The full version of this paper is available at [DKM19].

Our approach establishes a general-purpose reduction from minimax rates for sequential probability assignment for smoothed adversaries to minimax rates for transductive learning. This leads to optimal (logarithmic) fast rates for parametric classes and classes with finite VC dimension.