Our digital society is becoming more and more integrated facilitated by advancements in communication technology and services connecting our society from financial transactions to social interactions.
As a result, large quantities of data on our interactions, which can be used to quantitatively model and predict the evolution of social systems, are being generated on a daily basis. Mathematically, these
data can be characterized in terms of large and multiplex networks of how interactions between entities evolve in time forming so-called dynamic networks. As the digitalization of our interactions are disrupting practically all aspects of society from the financial sector to how information is shared, it is important to understand these complex dynamical systems of interactions and be able to foresee their behaviors. This is the focus of the proposed project: How can we develop and utilize large scale dynamic network embeddings to i) enable a human understanding of the structure of these complex systems and ii) forecast their future behaviors?
We also want to address the foundational question on the role of social for life outcomes in earnest. We address the role of social networks (family, friends, colleagues ) in stages. Firstly, we explore how
well can we predict life outcomes from features based on Danish registry data and network features. Here we explicitly compare the relative magnitude of the contributions from network and structural features. Third, we bring the network view into the deep learning paradigm. Here, the main idea is to rely on graph embeddings that incorporate nodal information What is the network’s contribution in predicting life outcomes? How do we incorporate deep nodal information and network evolution into graphembeddings?