1: \begin{abstract}

2:   In this paper, we aim to design and analyze distributed Bayesian

3:   estimation algorithms for sensor networks. The challenges we 

4:   address are to (i) derive a distributed provably-correct algorithm 

5:   in the functional space of probability distributions over continuous 

6:   variables, and (ii) leverage these results to obtain new distributed 

7:   estimators restricted to subsets of variables observed by individual 

8:   agents.  This relates to applications such as cooperative localization 

9:   and federated learning, where the data collected at any agent depends 

10:   on a subset of all variables of interest.  We present Bayesian 

11:   density estimation algorithms using data from non-linear likelihoods

12:   at agents in centralized, distributed, and marginal distributed

13:   settings.  After setting up a distributed estimation objective, we

14:   prove almost-sure convergence to the optimal set of pdfs at each agent. 

15:   Then, we prove the same for a storage-aware algorithm estimating 

16:   densities only over relevant variables at each agent. Finally, we 

17:   present a Gaussian version of these algorithms and implement it in 

18:   a mapping problem using variational inference to handle non-linear 

19:   likelihood models associated with LiDAR sensing.

20: \end{abstract}

21: