CITCoM

Causal Inference for Testing Computational Models

What is CITCoM and what is it aiming to achieve?

CITCoM is a 3.5 year research project funded by the EPSRC, running from early 2021 until mid 2024.

Computational models are becoming increasingly pervasive and important. This was illustrated in recent months, where government responses to the COVID pandemic have been largely informed by computational models. Clearly, it is important that these models are "correct" - incorrect outputs could potentially have disastrous consequences.

Computational models are commonly written in conventional programming languages, but are often harder to test than conventional systems for various reasons. For example, they tend to require large numbers of parameters, can take a long time to execute, and are often exploratory in nature, making it difficult to discern whether an output is to be expected or not.

To work around these problems, CITCoM will investigate the application of a family of statistical analysis approaches known as Causal Inference. These approaches provide a framework within which to select suitable parameter configurations, and to draw justifiable conclusions about model behaviour from small samples of executions (or simulations). There is even the possibility for counterfactual reasoning - answering "what-if" questions about executions that have not yet been executed.

CITCoM will use these capabilities to develop powerful new tools and methodologies, which can be used by modelling practitioners to rigorously test their computational models.

Who is involved?

News

  • "Deep State Inference: Toward Behavioural Model Inference of Black Box Software Systems" by Foozhan Ataiefard, Mohammad Mashhadi, Hadi Hemmati and Neil Walkinshaw, accepted to IEEE Transactions on Software Engineering. Preprint available here.

  • "Maternal hemodynamics and neonatal birthweight in pregnancies complicated by gestational diabetes: New insights from a novel causal inference analysis modelling" by Abi Anness, Andrew Clark, Kess Melhuish, Francesca Leone, Waseem Osman, Neil Walkinshaw, Asma Khalil, Thompson Robinson, David Webb, Hatem Mousa, published in the Journal of Ultrasound in Obstetrics and Gynecology, 2022. Paper available here.

  • "Test case generation for agent-based models: A systematic literature review" by Andrew G. Clark, Neil Walkinshaw and Rob Hierons accepted to the Journal of Information and Software Technology (IST). Preprint available here. (10/3/2021)