Assistant Professor of Biostatistics, UCLA
Talk Title
MCMC with Multiple Proposals
Abstract
Traditional MCMC algorithms generate a single proposal at every iteration, but there is no reason this must be the case. Indeed, the use of multiple proposals presents the opportunity for within-chain parallelization in ways that leverage high-performance conventional and quantum computing. I discuss the strange history of multiproposal MCMC algorithms before presenting a few recent advances, both practical and theoretical in nature. Finally, I outline directions of future research, including the bizarre benefits of bias.