I'm a PhD student in the Machine Learning Group at Cambridge, supervised by Rich Turner. I'm generally interested in probabilistic modelling and (approximate) inference, how Gaussian processes should feature in probabilistic programming, and how to scale GPs for large time series and spatio-temporal problems.
The above feeds directly into my work on probabilistic machine learning in climate science, which addresses combining the predictions of ensembles of GCMs in a sensible way, and the requirements that this task places on statistical weather modelling.
Gaussian process probabilistic programming (GPPP) is a term I've coined for the work I'm doing to re-design the way that we work with GPs in a practical sense. I presented the high-level aspects of this work at ProbProg and JuliaCon, and continue to work on a paper.
GPAR is a multi-output Gaussian process-based regression model. It's simple, scalable by GP standards, and seems to work really well in practice. In particular it overcomes some of the limitations of standard multi-output GPs. It appeared at AISTATS 2019.
The OILMM is another multi-output Gaussian process model for regression that's easily able to handle a lot of outputs while retaining exact inference. It's really Wessel's thing -- he came up with the main idea, and he and Eric did most of the work. I pointed out the connection with separable spatio-temporal processes and state space approximations, some ideas for experiments, and used TemporalGPs.jl to run one of the experiments. It appeared at ICML 2020.
This is a Julia implementation of my GPPP work. It's permanently ongoing, but makes working with GPs in problems involving multiple related processes signficantly more straightforward than traditional GP packages.
This work implements SDE approximations to GPs, which dramatically accelerates inference and learning for models involving long time horizons. This is notable because the standard pseudo-point approximations fail in these scenarios.