Title: Metropolis-within-piecewise deterministic Markov processes
Authors: Kengo Kamatani - ISM (Japan) [presenting]
Naohisa Okamoto - Osaka University (Japan)
Abstract: The Metropolis-Hastings (MH) jump update to the piecewise deterministic Markov processes (PDMP) is presented. PDMPs are useful for Bayesian computation, allowing unbiased subsampling. However, the implementation of PDMP requires efficient coding of the jump times. By combining the MH scheme, we can bypass this difficulty for some parameters of interest. We will show some theoretical properties of Markov processes defined by Metropolis-within-PDMPs. Finally, we apply it to Bayesian inference for stochastic processes.