1: \begin{abstract}

2: In this paper, we are concerned with estimating the joint probability

3: of random variables $X$ and $Y$,  given $N$ independent observation blocks

4: $(\zb x^i,\zb y^i)$, $i=1,\ldots,N$, 

5: each of $M$ samples 

6: $(\zb x^i,\zb y^i) = \big( (x^i_j, y^i_{\sigma^i(j)}) \big)_{j=1}^M$,

7: where $\sigma^i$ denotes an unknown permutation 

8: of i.i.d.~sampled pairs

9: $(x^i_j,y_j^i)$, $j=1,\ldots,M$. 

10: This means that the internal ordering 

11: of the $M$ samples within an observation block is not known. 

12: We derive a maximum-likelihood inference functional, propose a computationally tractable approximation and analyze their properties. In particular, we prove a $\Gamma$-convergence

13: result showing that we can recover the true density 

14: from empirical approximations as the number $N$ of blocks goes to infinity.

15: Using entropic optimal transport kernels, we model a class of  hypothesis spaces of density functions over which the inference functional can be minimized.

16: This hypothesis class is particularly suited for

17: approximate  inference of transfer operators from data. 

18: We solve the resulting discrete minimization problem by 

19: a modification of the EMML algorithm to take addional transition probability

20: constraints into account and prove the convergence of this algorithm.

21: Proof-of-concept examples demonstrate the potential of our method.

22: \end{abstract}

23: