Topic: Identifying Epidemic Dynamics from Data using Machine Learning
Description: Epidemic dynamics can be characterised by the rates at which individuals move between different health states, such as rates of infection, recovery, hospitalisation, and death. The numbers of people transitioning between different states per unit time are known as the fluxes. In traditional epidemic modelling, the mathematical form of these fluxes is typically assumed, which can prevent models from accurately capturing complex real-world dynamics.
This project aims to develop novel methods for learning the mathematical equations underlying the fluxes directly from observational data. By learning these dynamics directly from data, we avoid imposing restrictive assumptions and better reflect reality. Learning mathematical expressions for the fluxes also ensures interpretability, which is key for policymakers who rely on epidemic models to make informed public health decisions.
This project will develop system identification methods that address the challenges associated with epidemic data, such as noisy and partially observed data. While the methods will be developed for epidemic systems, they will be applicable to any dynamic system where entities flow into and out of defined states, offering a generalisable approach to dynamical system identification.
Supervisors:
Lancaster University: Dr Lloyd Chapman, Dr Christopher Jewell
University of Oslo: Dr Arnoldo Frigessi, Dr Alvaro Kohn Luque
IndabaX South Africa 2023
Talk: Contrastive Learning: Fundamentals and Practical Applications
Introduction to supervised and self-supervised contrastive learning and applications of these methods across computer vision, natural language processing and recommender systems
Research Project: Automated Malignant Melanoma Detection using Supervised Contrastive Learning
Key words:
Melanoma detection; supervised contrastive learning; computer-aided detection
Recommendation System Project: Recommender systems on MovieLens data using explicit ratings, and curated implicit feedback data.
Natural Language Processing Project: Assessing word embedding evaluation methodologies using the AG’s News Topic Classification Dataset
Research Report: High-dimensional Co-occurrence Modelling with an Application in Disease Co-morbidity
Key words:
Co-occurrence modelling; multi-morbidity; dependent binary simulation; spectral clustering
