Th e use of Bayesian methods in large-scale data sett ings is attractive because of the rich hierarchica l models\, uncertainty quantification\, and prior specification they provide. However\, standard Bay esian inference algorithms are computationally exp ensive\, so their direct application to large data sets can be difficult or infeasible. Rather than m odify existing algorithms\, we instead leverage th e insight that data is often redundant via a pre-p rocessing step. In particular\, we construct a wei ghted subset of the data (called a coreset) that i s much smaller than the original dataset. We then input this small coreset to existing posterior inf erence algorithms without modification. To demonst rate the feasibility of this approach\, we develop an efficient coreset construction algorithm for B ayesian logistic regression models. We provide the oretical guarantees on the size and approximation quality of the coreset -- both for fixed\, known d atasets\, and in expectation for a wide class o f data generative models. Our approach permits effic ient construction of the coreset in both streaming and parallel settings\, with minimal additional e ffort. We demonstrate the efficacy of our approach on a number of synthetic and real-world datasets\ , and find that\, in practice\, the size of the co reset is independent of the original dataset size. LOCATION:Seminar Room 1\, Newton Institute CONTACT:INI IT END:VEVENT END:VCALENDAR