B1163
Title: MCMC inference for discretely-observed diffusions: Improving efficiency
Authors: Christiane Fuchs - Helmholtz Center Munich (Germany) [presenting]
Susanne Pieschner - Helmholtz Center Munich (Germany)
Abstract: Diffusion processes are used to realistically model time-continuous processes in biology. In real-data applications, inference for diffusions is difficult when the likelihood function is analytically unavailable. A widely applicable method for parameter estimation is a Markov chain Monte Carlo (MCMC) approach which simulates and employs synthetic sample paths. For high-dimensional processes, however, the method is computationally intensive. We present attempts to improve estimation efficiency in practice, including approaches that did not turn out to be successful. We illustrate the inferential methods on time-resolved single-cell gene expression data.
Journal