PhD studentship in Uncertainty Quantification for Machine Learning at Ghent University (Belgium)
Project title: Development of new machine learning methods that distinguish aleatoric and epistemic uncertainty
Duration of studentship: 4 years with an evaluation after one year
Studentship start date: October 2022 or later
Application closing date: August 15th 2022 (will be extended if no suitable candidate is found). Apply as soon as possible to avoid disappointment!
Trustworthy machine learning (ML) systems should be able to express their uncertainty about the predictions they make, i.e., they should know what they don’t know. However, complex ML methods, such as deep neural networks, often fail to express their uncertainty in an honest way. This is a crucial problem in application domains such as medicine, climate and biology, because ML systems can only assist end users if one can trust the ML systems.
In this PhD project we intend to develop and evaluate several ML methodologies to quantify uncertainty. We want to explore four different ways of quantifying uncertainty, resulting in four different tasks. The first two approaches, in which distributions or sets are used to express uncertainty, respectively, are commonly used in supervised learning. Those approaches typically only take aleatoric uncertainty into account, i.e., uncertainty that cannot be reduced by collecting data of more patients. The last two approaches, which express a second-order uncertainty via distributions of distributions or sets of distributions, are less established in supervised learning. However, those approaches are gaining popularity, because they are able to represent epistemic uncertainty as well, i.e., uncertainty that arises because of working with limited sample sizes, and that can be reduced by collecting more data.
The position is available in the group of Prof. Willem Waegeman, at the Department of Data Analysis and Mathematical Modelling of Ghent University (Belgium). At this moment specific interests of the research group are deep learning, multi-target prediction, uncertainty quantification, sequence learning and time series analysis. The group is also engaged in applications that have a positive impact on our society, with a focus on domains such as molecular biology, medicine, environmental sciences, and climate analysis.
The ideal candidate for the position has the following profile:
- An MSc degree in Computer Science, Physics, Mathematics, Statistics, (Bio-)Engineering or equivalent – candidates from outside Belgium are welcome, but they are expected to move to Ghent for four years
- An interest in fundamental machine learning research, as well as practical applications in the life sciences
- Experience with machine learning and statistics
- Experience with programming in Python
- Fluent in English (speaking and writing)
- Team player with good communication skills
How to apply
Send your c.v., a short motivation, a copy of your MSc-thesis and any relevant publications to Mrs. Ruth Van Den Driessche (secretary of the department) Email: firstname.lastname@example.org. For further information about the position, you can contact Prof. Willem Waegeman directly.