Skip to main content

PhD Seminar: Laura Natali

Science and Information Technology

The Physics PhD Seminar Series is held over Zoom. One PhD student gives a 15-minute presentation on their work, followed by questions. The presentations are aimed at a general physics audience.

8 Mar 2021
12:00 - 12:30

Laura Natali


Improving epidemic testing and containment strategies using machine learning


The work we present deals with the containment of epidemic outbreaks and aims to improve the use of a limited amount of available resources. Cost-effective containment strategies rely on efficiently identifying infected individuals, making the best use of testing resources. We present an application of machine learning to improve testing strategies in the event of epidemic outbreaks.

We adopt a simplified model for the population based on the SIR model — first proposed in 1927 by Kermack and McKendrick — in which the population is split into three groups: Susceptibles, Infectious and Recovered/Removed. The groups respectively include individuals that have never been, currently are and have previously been infectious. In such a framework, the free evolution of the outbreak is compared to a model of contact tracing and the proposed neural-network-based strategy.

The neural network is trained on spatial and temporal data about the confirmed cases, to better elaborate the information available. The result for the epidemic evolution in time is significantly improved and the total fraction of the population infected is reduced by half by introducing the neural-network-informed strategy.

Bio: Laura Natali

I am at the end of my first year as a PhD student at the University of Gothenburg. My main supervisor is Giovanni Volpe, I am part of the soft matter lab and my office is on the ground floor of Soliden building. I am part of the ActiveMatter network, that includes several universities across Europe to investigate experimental, theoretical and computational aspects of systems out-of-equilibrium. My main main focus is on machine learning approaches to active systems