PhD project: Bayesian Optimisation & Priors — Improving Air Pollution Monitoring and Hyperparameter Optimisation

My PhD has two main focus areas: transfer learning for Bayesian optimisation, and its application to air pollution monitoring.

Transfer learning for Bayesian optimisation

Bayesian optimisation is an efficient way of optimising black-box functions. But the efficiency of the sampling depends on the quality of the underlying model. We’re exploring how to make Bayesian optimisation more sample-efficient by learning from related optimisation tasks.

Our work on transfer leraning for Bayesian optimisation was presented at workshops at AutoML 2023 and NeurIPS 2023. See Publications.

Air pollution monitoring

We’ve explored the use of Bayesian optimisation for urban air pollution monitoring. New lower-cost sensors mean many more locations can be sampled, but with higher uncertainty. We’re using Bayesian optimisation to model pollution levels and uncertainties in an area, and determine where to sample next. To tune the models we’re also using hierarchical modelling and Markov chain Monte Carlo.

Our work on air pollution monitoring was presented at AAAI 2022 and at the AI for Earth Sciences workshop at NeurIPS 2020. See Publications.