Jointly led by UGB and IRD, this research explores the optimization of epidemiological simulations through multi-agent systems.
The Nhan Ho, researcher at UMMISCO (Université Gaston Berger de Saint-Louis Sénégal).
Thesis defended in 2016 at the doctoral school École doctorale Informatique, télécommunications et électronique de Paris (1992-....).
This research is the result of an international co-supervision between several partner institutions.
Referenced in the ABES/STAR network, this thesis meets the rigorous criteria of French higher education.
The study of complex systems, such as ecosystems or urban dynamics, often relies on simulators. These tools, which can be seen as virtual laboratories, not only allow for understanding observed behaviors but also for anticipating the evolution of systems in various contexts. For example, in the field of epidemiology, these simulators can help predict the spread of a disease, assess the impact of different health interventions, or plan vaccination campaigns. However, trust in the results of a simulation depends on the quality of its validation. This validation requires a thorough analysis of the model, a sensitivity study of the parameters, and a confrontation with real data. In other words, it is not enough to simulate; it is also necessary to ensure that the results are reliable and relevant. In this context, the GRADEA protocol has been developed to optimize the exploration of parameter spaces of complex simulators.
The GRADEA protocol combines three exploration algorithms: screening search, global search, and local search. Each of these algorithms has its own strengths and weaknesses, but together they form a synergistic approach. Screening search allows for quickly reducing the search space by identifying the most influential parameters. Global search, on the other hand, explores the entire parameter space, ensuring that potential solutions are not overlooked. Finally, local search refines the results by focusing on promising areas identified by the first two methods. These algorithms operate in parallel, allowing for the identification of areas of interest and effectively mapping the solution space. This approach has been successfully applied to an environmental simulator for planning vaccination policies against measles in Vietnam. Preliminary results show a significant improvement in the efficiency of parameter exploration, which could have major implications for public health. Indeed, a better understanding of epidemiological dynamics could allow for more effective targeting of resources and optimization of health interventions.
The modular agent-based architecture allows interaction with computing nodes to execute simulations, thereby optimizing the use of high-performance computing resources. This offers researchers the opportunity to combine algorithms tailored to their specific needs while leveraging available resources to accelerate exploration. For example, in a context where computing resources are limited, this modularity allows for distributing simulation tasks across multiple machines, thus increasing the speed and efficiency of analysis. This is particularly relevant in regions like West Africa, where research infrastructures may be less developed, but the need for precise and rapid analyses is crucial.
It is crucial to ask how this approach could be adapted to other fields of study, particularly in West Africa, where complex systems are ubiquitous. Policymakers must consider the integration of such technologies into their public health and land-use strategies. For example, modeling water supply systems or agricultural dynamics could benefit from the GRADEA approach, allowing for the anticipation of food crises or water shortages. Indeed, the ability to model and anticipate the dynamics of complex systems could transform the way policies are designed and implemented. This could also foster better collaboration between researchers, policymakers, and local communities, ensuring that policies are based on evidence and reliable simulations.
In summary, the GRADEA protocol represents a significant advancement in optimizing epidemiological simulations. It paves the way for a better understanding of complex systems and more effective planning of public policies. By integrating optimization algorithms within a multi-agent framework, it not only improves the accuracy of simulations but also accelerates the solution exploration process. In the future, it will be essential to continue exploring these synergies between technology and research to address the complex challenges our societies face.
Access the full study
Sources and Access
The Nhan Ho. Couplage d'algorithmes d'optimisation par un système multi-agents pour l'exploration distribuée de simulateurs complexes : application à l'épidémiologie. Calcul parallèle, distribué et partagé [cs.DC]. Université Pierre et Marie Curie - Paris VI, 2016. Français. ⟨NNT : 2016PA066547⟩. ⟨tel-01531905⟩