Statistical Machine Learning for Dynamic Network Modelling

Descripción del proyecto

Modelización dinámica de redes para la toma de decisiones

El proyecto DyNeMo tiene por objeto desarrollar modelos estadísticos y de aprendizaje automático avanzados para comprender y predecir redes complejas en evolución. Los modelos se aplicarán a dos conjuntos de datos únicos para responder preguntas científicas relevantes con repercusión social y ambiental, y ayudar a influir en las políticas públicas. La primera pregunta se centra en cómo las empresas de todo el mundo pueden mejorar su repercusión socioambiental de acuerdo con los Objetivos de Desarrollo Sostenible de las Naciones Unidas. La segunda evalúa y predice el comportamiento de las averías en cascada de tuberías en los sistemas de distribución de agua. El equipo del proyecto DyNeMo, que cuenta con el respaldo de las acciones Marie Skłodowska-Curie, espera que sus resultados ofrezcan predicciones relevantes sobre la repercusión socioambiental para facilitar la toma de decisiones.

Objetivo

Networks are ubiquitous in the society, technology, biology and economy. Being able to perform accurate inference and prediction on such data is highly non-trivial due to their large size and complex dynamic behaviour. In many networks such as social ones, connectivity patterns and network structurechange dynamically. On the other hand, technological networks such as water systems have fixed structure but dynamic processes might take place on them appear, such as failures. DyNeMo is a cross-disciplinary project that aims to provide a novel, explainable yet scalable framework using statistical machine learning to model dynamic behavior on networks, with the goal to answer crucial scientific questions with social and environmental impact, influencing public policy. The project combines statistical tools that ensure flexibility, data adaptivity and interpretable parameters with Deep Learning tools ensuring efficiency and scalability. The methods will be used to answer two highly impactful scientific questions given access on unique datasets. The first is how companies worldwide can maximize their socio-environmental impact aligned with UN sustainable goal strategies. The second question is how to quantify and predict behaviour of cascading pipe failures on the water distribution systems in Cyprus. The results will detect patterns and perform prediction, thus influencing public policy in alignment to the EU strategic goals on monitoring water systems. The fellow and the supervisor have complementary expertise which in combination with the plan, the arrangements and environment provided by the host institution ensure a successful implementation of this novel and timely project as well as an effective dissemination and utilisation of the expected outcomes. The results of the proposed action are expected to have important conclusions on socio-environmental impact and can shape decision making and policy making in the EU and around the world.