Statistical and machine learning models developed in academic settings rarely translate into clinical or engineering practice. Practitioners in safety-critical sectors are often reluctant to adopt new methods, and this scepticism is valid: they are asked to carry professional and legal liability for decisions informed by models whose risks have not been adequately quantified. Meanwhile, academic incentives reward novelty over implementation, and models are frequently developed without clear links to the decisions they are meant to support.

 

In this seminar, I propose approaches to two key barriers in translating models from research to practice. First, I introduce "maximally communicative" modelling to improve engagement with decision-makers and ensure that model outputs are explicitly tied to actionable decisions. Second, I discuss model risk management, drawing on lessons from the financial sector where regulatory frameworks emerged following the 2008 crisis, and show how these principles can be adapted for engineering and healthcare contexts.

iCalendar URL:
TBC
How to use our iCalendar feeds
Download current list of all events in the series (will not update)