multivariate time series forecasting models for infectious diseases: deep learning, multimodal signals, and uncertainty

This project will aim to develop state-of-the-art multivariate time series forecasting models for tracking the prevalence of infectious diseases such as influenza and COVID-19. The core methodological contributions will focus on the development of time series forecasting models that will be capable of: (a) learning from multimodal signals (from epidemiological endpoints to online user activity), (b) combining optimisation functions from state-of-the-art deep learning models and more conventional statistical approaches in epidemiology, (c) accurately modelling prediction uncertainty, and (d) being directly deployable by interested stakeholders. Outcomes will be incorporated in syndromic surveillance systems used by UK's Health Security Agency (UKHSA) as well as more broadly, similarly to our flu and COVID-19 models (see fludetector.cs.ucl.ac.uk and covid.cs.ucl.ac.uk).