Inference in complex systems
The inference in complex systems research area develops new techniques to model high-dimensional data from different scientific domains.
The size and complexity of modern experimental observations has created a data-interpretation bottleneck in many areas of scientific research.
'Black-box’ machine learning techniques are often not appropriate for addressing research questions. These techniques emphasise prediction at the expense of interpretability and are difficult to apply to high-dimensional data with small sample sizes.
To address these problems, we design techniques with domain-specific constraints on symmetry and sparsity. This results in models that are more interpretable and suitable to scientific inquiry.
Our approach to research is:
- experimental: addressing domain-specific research questions by developing interpretable models of high-dimensional data.
- theoretical: exploring phase transitions in statistical inference and their consequences on information processing and decision-making.
- educational: nurturing a new generation of researchers who can think about high-dimensional data and work with domain specialists across disciplines.
Read about our current research projects.
Variation in cell lineages
This project explores ways to identify the key stages of development in a cell lineage where phenotypic noise is high. An inference technique has been designed that exploits the group symmetry of trees and can be applied to variation in any tree structure.
3D lineage tracking
This project involves developing an accurate and efficient way of extracting cell phenotypes and lineage relationships directly from high-resolution microscopy movies of proliferating cell populations.
Neurocortical model inference
This project involves parameter estimation in dynamical systems models. In particular, a mechanistic model of neurocortical activity is being used to detect physiologically-relevant predictors of an individual’s response to treatments such as anaesthesia.
Area leader: Professor Damien Hicks