Hirt, Marcel Andre; (2021) Approximate inference methods in probabilistic machine learning and Bayesian statistics. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text PhD_Thesis_Marcel_Hirt_final.pdf Download (10MB) | Preview

Abstract

This thesis develops new methods for efficient approximate inference in probabilistic models. Such models are routinely used in different fields, yet they remain computationally challenging as they involve high-dimensional integrals. We propose different approximate inference approaches addressing some challenges in probabilistic machine learning and Bayesian statistics. First, we present a Bayesian framework for genome-wide inference of DNA methylation levels and devise an efficient particle filtering and smoothing algorithm that can be used to identify differentially methylated regions between case and control groups. Second, we present a scalable inference approach for state space models by combining variational methods with sequential Monte Carlo sampling. The method is applied to self-exciting point process models that allow for flexible dynamics in the latent intensity function. Third, a new variational density motivated by copulas is developed. This new variational family can be beneficial compared with Gaussian approximations, as illustrated on examples with Bayesian neural networks. Lastly, we make some progress in a gradient-based adaptation of Hamiltonian Monte Carlo samplers by maximizing an approximation of the proposal entropy.

Download activity - last month

Export as JSONCSVXML

Download activity - last 12 months

Export as JSONCSVXML

Downloads by country - last 12 months

Export as XMLCSVJSON

Archive Staff Only

View Item