Speaker: Sam Livingstone, University College London

Abstract: I will discuss some recent work on gradient-based Markov chain Monte Carlo. At their best gradient-based algorithms are often state-of-the-art, but they can also behave erratically in various ways.  In the first part of the talk I’ll introduce a gradient-based scheme called the Barker proposal that can often match the performance of competitors such as MALA and HMC but is probably more robust. I will in particular highlight the benefits this brings in the context of adaptive MCMC.  If time permits, in the second part of the talk I will talk about ongoing using MCMC for automatic model selection in poly-hazard models.  The first part is joint work with Giacomo Zanella, Jure Vogrinc & Max Hird, the second is joint with Luke Hardcastle and Gianluca Baio.

This will be a free hybrid seminar. To register to attend virtually, please click here: https://us02web.zoom.us/meeting/register/tZUsdOypqTwrH9LeiEEIL8dXxO_IpOC8lVMS