ESR11 – Data-driven support to understanding of complex dynamical physical phenomena, such as epidemics

ESR11: Data-driven support to understanding of complex dynamical physical phenomena, such as epidemics

Recruiting beneficiary: University of Torino, Italy


Internal supervisors: Prof. Maria Luisa Sapino, Prof. Matteo Sereno


Brief project description: In this project, the ESR will: (i) develop a learning framework to assist decision makers, suitable for the complicated dynamics of the systems interconnected under epidemic scenarios and the highly heterogeneous systems, varying in spatial and temporal scales data; (ii) improve and refine the above framework, to take into account the peculiarities of complex natural and human-based systems, such sparse and noisy observations; (iii) develop efficient sampling algorithms that, with limited available simulation budget (and therefore inherently sparse model results), allow to capture the main characteristics of the dynamically evolving system of interest.


Updates: Currently, Javier is working on a parameter sampling method that can be applied to a set of models to independently sample for each of the given models while preserving the commonalities they have. In addition, and related to the above, he is also working on a pivot guided mechanism to identify simulation instances to execute next as the state of the simulation changes for future sample selection. These two works will make it possible to sample very complex models with many parameters in an efficient way, obtaining very accurate results without an unbearable computational cost.
During his secondments, Javier has also applied Bayesian methods to calibrate a complex compartmental epidemiological model and has developed an optimization algorithm to support exploration of alternative potential timelines within the continuous-coupled ensemble simulation framework, DataStorm.


Selected contributions:

Candan, K.S., Sheth, P., Mandal, P., Azad, F.T., Li, M-L. , Arslan, B., Muenich, R., Liu, H., Chowell-Puente, G., Sabo, J., Redondo Anton, J., & Sapino, M. L. A Vision for Spatio-Causal Situation Awareness, Forecasting, and Planning. (Submitted 2023) ACM Transactions on Spatial Algorithms and Systems.


Redondo Anton, J.. The Multi-Armed Bandit Problem. In Machine Learning methods for data analysis, Maths Volunteers, September 2023. [information & video].