Program Leader: Luk Peeters
The program focuses on conceptualisation; the formulation of an internally consistent set of assumptions and hypothesis that are able to explain the available information about the system under study. Such conceptualisation underpins both physically-based and data-driven quantitative models for inference and prediction. The results of quantitative models are therefore always conditional on the assumptions implicit in the conceptualisation. This research program aims to make these assumptions explicit so that they can be accounted for in decision making under uncertainty.
The program aligns principally with the conceptualisation pillar but is strongly linked to the imaging and prediction pillars. Imaging is essential for exploratory data analysis – the critical examination, visualisation and correlation of a wide variety of data sets is integral when creating a conceptual model of a system. The System Conceptualisation program focuses on providing a formal framework to capture domain knowledge from the interpretation of data and provide strategies to systematically formulate alternative interpretations of the available data.
The prediction pillar, on the other hand, codifies the conceptualisation and allows the hypotheses of the conceptualisation to be tested. The program aims to develop a formal causal basis to verify that predictive models give the correct answers for the right reasons. This is especially important for machine learning methods, where a collection of data-driven techniques are used to implicitly define a statistical relationship in order to predict unknown quantities. The program will work to develop machine learning approaches that will allow system understanding to be integrated and accounted for.