Our paper on Bayesian Coresets: Revisiting the Nonconvex Optimization Perspective is accepted at the AISTATS conference this year (virtual).

• Bayesian Coresets: Revisiting the Nonconvex Optimization Perspective:

Abstract. Bayesian coresets have emerged as a promising approach for implementing scalable Bayesian inference. The Bayesian coreset problem in- volves selecting a (weighted) subset of the data samples, such that the posterior inference us- ing the selected subset closely approximates the posterior inference using the full dataset. This manuscript revisits Bayesian coresets through the lens of sparsity constrained opti- mization. Leveraging recent advances in ac- celerated optimization methods, we propose and analyze a novel algorithm for coreset se- lection. We provide explicit convergence rate guarantees and present an empirical evalua- tion on a variety of benchmark datasets to highlight our proposed algorithm’s superior performance compared to state-of-the-art on speed and accuracy.