Bayesian variable selection in high-dimensional latent variable models

The project will develop new methodological and computational tools for performing efficient Bayesian variable selection in hierarchical models that include large numbers of latent variables. The successful candidate will build on recent work by the supervisory team [1][2] and will extend the work in a number of directions, developing sophisticated and efficient algorithms for fast computation in complex models, while developing associated tools, such as R packages, to ensure dissemination of the methods. An example application of the work is the recent increase in availability of satellite remote-sensing imagery to measure the environment in detail [3], which gives rise to hundreds or thousands of covariates that, in combination with sophisticated statistical models for ecological data, can revolutionise landscape and conservation planning. [1] https://discovery.ucl.ac.uk/id/eprint/10098647/ [2] https://rss.onlinelibrary.wiley.com/doi/abs/10.1111/rssc.12390?af=R [3] https://www.nature.com/articles/s41559-017-0176