Research exchange with the VUB AI group and SIMID

What I work on and who with, composable EpiAware modelling, recent delays work, and where AI might plug in. A morning of short talks from the VUB Artificial Intelligence group and the UHasselt and UAntwerp SIMID group, at the AI Experience Centre, Brussels.

Research exchange, VUB AI group and UH-UA SIMID group · Tuesday 15 September 2026, 09:45–10:15 (Europe/Brussels) · AI Experience Centre, Pleinlaan 9, 1050 Elsene · hybrid · Sam Abbott

Slides epinowcast EpiAware Google Scholar

Abstract

I work at the London School of Hygiene & Tropical Medicine, in the Centre for Mathematical Modelling of Infectious Diseases. Some of my main research areas are improving outbreak modelling through composable models, where components can be reused across contexts, and a workflow for modelling with multiple data sources. I am a member of the epiforecasts group led by Sebastian Funk, whose shared focus is broadly real-time modelling and evaluating it, through packages such as EpiNow2 and scoringutils and through the forecast hubs. I created the epinowcast community, which has a forum, a monthly seminar series and a set of packages for nowcasting and delay estimation. The composable modelling work is currently in Julia, in the EpiAware organisation. This organisation is loosely an extension of the epinowcast community but is in its initial stages. I have also spent time consulting with collaborators at the US CDC, UKHSA, and state and local public health departments in the US. The tools from this work are widely used in outbreak response.

The talk then covers three strands of current work. Composable modelling in EpiAware, the two approaches we are taking to it, and the AI methods that might plug into it. The workflow we think models should be built and checked with, and where agents sit in it. Recent work on epidemiological delays and the censoring and truncation biases in estimating them, in primarycensored, epidist and CensoredDistributions.jl.

Where AI might plug in is an open question. I flag some potentially exciting approaches, including universal differential equations, physics-informed neural networks, a renewal process as a neural network layer, agents building and checking components, and an outbreak-specific foundation model.

The organisers proposed a research exchange the day after the workshop, hosted by Pieter Libin’s AI Lab at VUB, at which selected members of the VUB AI group and the UHasselt and UAntwerp infectious disease modelling group share their work. They asked me to talk about my background, research group and current work, for 20 to 25 minutes with questions after. There is no theme for the day.

The day before

Resources

Groups I work with

  • epiforecasts. The group’s GitHub organisation.
  • epinowcast. Nowcasting and delay estimation, with a community forum and seminar series.
  • EpiAware. Composable infectious disease modelling in Julia.

Current work