1: \begin{abstract}

2: A Bayesian coreset is a small, weighted subset of data that replaces the

3: full dataset during inference in order to reduce computational cost.

4: However, state of the art methods for tuning coreset weights are

5: expensive, require nontrivial user input, and impose constraints on the model.

6: In this work, we propose a new method---\emph{Coreset MCMC}---that 

7: simulates a Markov chain targeting the coreset posterior, while simultaneously updating

8: the coreset weights using those same draws. Coreset MCMC 

9: is simple to implement and tune, and can be used with any existing MCMC kernel. 

10: We analyze Coreset MCMC in a representative setting to obtain key insights about the convergence

11: behaviour of the method. Empirical results demonstrate that Coreset MCMC

12: provides higher quality posterior approximations and reduced computational cost

13: compared with other coreset construction methods. Further, compared with other 

14: general subsampling MCMC methods, we find that Coreset MCMC has a higher sampling 

15: efficiency with competitively accurate posterior approximations.

16: \end{abstract}

17: