1: \begin{abstract}

2:     Regularized optimal transport (OT) 

3:     has received much attention in recent years

4:     starting from Cuturi's  paper with Kullback-Leibler (KL) divergence regularized OT.

5:     In this paper, we propose to regularize the OT problem

6:     using the family of $\alpha$-Rényi divergences for $\alpha \in (0, 1)$.

7:     Rényi divergences are neither $f$-divergences 

8:     nor Bregman distances, 

9:     but they recover the KL divergence in the limit $\alpha \nearrow 1$.

10:     The advantage of introducing the additional parameter $\alpha$ is 

11:     that for $\alpha \searrow 0$ we obtain convergence to the unregularized OT problem.

12:     For the KL regularized OT problem, this was achieved by letting the regularization parameter $\frac{1}{\lambda}$ tend to zero, 

13:     which causes numerical instabilities.

14: 

15:     We present two different ways to obtain premetrics on probability measures, namely

16:     by Rényi divergence constraints and penalization.

17:     The latter premetric interpolates

18:     between the unregularized and KL regularized OT problem 

19:     with weak convergence of the minimizer, 

20:     generalizing the interpolating property of KL regularized OT.

21: 

22:     We use a nested mirror descent algorithm for solving the primal formulation.

23:     Both on real and synthetic data sets 

24:     Rényi regularized OT plans outperform their KL and Tsallis counterparts 

25:     in terms of being closer to the unregularized transport plans 

26:     and recovering the ground truth in inference tasks better.

27: \end{abstract}

28: