Fully funded studentship Modern Statistics and Statistical Machine Learning Centre for Doctoral Training
Due to funding constraints to be eligible for this studentship you need to be classified as a home student. Please see eligibility criteria here: Eligibility for studentship funding – UKRI
Deadline for Applications: Wednesday 31 May at 5.00 pm
Academic Supervisors: Calvin Tsay and Mark van der Wilk (Imperial).
Contact: c.tsay@imperial.ac.uk
Project Title: Meta-Learning for Bayesian Optimization in Experimental Campaigns
This project investigates meta-learning in the context of designing experimental campaigns via Bayesian optimisation. In this context, experiments can be relatively expensive, and we seek to improve their design (i.e., data acquisition strategies) by meta-learning from existing experimental campaigns. These may share characteristics such as (i) how the input-output relationship affects optimal sampling, (ii) how a constrained input space should be explored, and (iii) how different types (e.g., multi-fidelity) of output measurements should be utilized.
Towards this goal, and noting the above emphasis on data efficiency, we will explore methodologies at the intersection of Bayesian optimisation, meta-learning, Gaussian processes, and/or learning invariances/symmetries from data. The student will ideally have an interest in contributing to open-source software packages. This project is part of a larger research collaboration between Imperial and BASF.
The successful candidate will be part of the Modern Statistics and Statistical Machine Learning Centre for Doctoral Training and will be required to take courses and attend events as part of the cohort.
To apply please send an up-to-date CV to statml.io.admissions@imperial@ac.uk with the subject “Meta-Learning for Bayesian Optimization in Experimental Campaigns”