Senior Research Officer in Machine Learning

The University of Essex offers an exciting opportunity for a Senior Research Officer in Machine Learning to work on out-of-distribution generalisation for tabular data, under the EPSRC-funded project “Element-compositor methods for out-of-distribution machine learning”.

The Institute for Analytics and Data Science (IADS) was created in July 2014 to bring together all the University’s expertise and work in analytics and data science. IADS research utilises new forms of data and advanced analytic techniques to tackle social challenges, engage with business and the wider non-academic community, is international in outlook and scope, and multidisciplinary and interdisciplinary by nature.

As it stands, machine learning (ML) methods are brittle. If there is a data distribution shift between the training and test set, most methods tend to fail, often unpredictably, as they cannot extrapolate. For example, if one has only collected crop data during relative cold periods, one abnormally warm period tends to confuse most models. These "distribution shifts" are often not visible, but omnipresent, as in higher dimensions it is not easy to discern where the data gaps lie. The ability to extrapolate from data successfully is often called "out-of-distribution" (OOD) generalisation. For this post, we are interested in attracting a researcher working on fundamental tabular ML OOD problems, with a focus on developing new methods and techniques that will improve standard ML algorithms so as to achieve good performance outside their training distribution, by treating high-dimensional problems as an explicit composition of low-dimensional latent features. Our objectives are as follows:

• Establish a complete evaluation framework for OOD generalisation on tabular data, which includes standard supervised learning, causal learning, and off-policy reinforcement learning.
• Improve our understanding of how big the distributional shift should be before generalisation becomes impossible for mainstream algorithms.
• Develop new methods for OOD generalisation based on the number and type of compositions and basic feature functions that are meaningful in real-world data.
• Develop refutation methods to help identify whether a composition identified is sensible.

You should hold a PhD in Artificial Intelligence or relevant numerate topic (or be nearing completion). You will have a publication record appropriate for your experience and be able to demonstrate evidence of potential to develop an internationally leading profile in your field. Strong written and verbal communication skills are also essential.

A full list of duties and responsibilities can be found within the jobpack.

This post is fixed-term until 14 September 2028.

As part of the process and to be successful you will be required to demonstrate how you your research has been impactful within and beyond the academic sphere, recognising the importance of this and the effect it will have on the University.

The University of Essex is a campus-based University and all academic appointments are made with the expectation that you will be present on campus to fulfil the requirements of your position. At the time of a job offer, the possibility of hybrid working, will be discussed and considered on an individual basis, but you will be required to be based within the UK.

Advert information