1: \begin{abstract}

2:   We introduce the Conic Blackwell Algorithm$^+$ (\cbap) regret minimizer, a new parameter- and scale-free regret minimizer for general convex sets.

3:   \cbap{} is based on Blackwell approachability and attains $O(\sqrt{T})$ regret.

4:   We show how to efficiently instantiate \cbap{} for many decision sets of interest, including the simplex, $\ell_{p}$ norm balls, and ellipsoidal confidence regions in the simplex.

5:   Based on \cbap\, we introduce \spcbap\, a new parameter-free algorithm for solving convex-concave saddle-point problems, which achieves a $O(1/\sqrt{T})$ ergodic rate of convergence.  In our simulations, we demonstrate the wide applicability of \spcbap{} on several standard saddle-point problems, including matrix games, extensive-form games, distributionally robust logistic regression, and Markov decision processes. In each setting, \spcbap{} achieves state-of-the-art numerical performance, and outperforms classical methods, without the need for any choice of step sizes or other algorithmic parameters.

6: \end{abstract}

7: