Aleksi

Winstén

With the rapid growth of large-scale information networks, understanding and predicting time-evolving relationships in temporal networks has become increasingly important across a wide range of application areas. A prominent approach for representing relational data is provided by latent position network models, which embed nodes into a euclidean space in order to capture network structure and enable visualization, inference, and prediction.

While latent position models and modern embedding methods have achieved strong empirical success for large static networks, temporal network prediction remains challenging because node interactions evolve over time and the resulting models can be computationally demanding. Developing scalable models for network evolution is crucial, since it would enable forecasting of macroscopic social phenomena such as community formation and polarization.

My research addresses these challenges by developing a physics-inspired framework for temporal network modeling. The central idea is to derive a potential function for the latent space of a network and interpret network evolution with it. This approach builds on our recently developed notion of internal time in networks, which provides an implicit force-field structure for the latent dynamics. The motivation is computational as well as conceptual: numerical methods for simulating N-body dynamics in force fields can be substantially faster than many existing temporal embedding techniques, offering a promising route toward scalable prediction and simulation of evolving networks.

In parallel, I conduct applied research in anticoagulation therapy for atrial fibrillation. My interest in this area is the mathematical modeling of patient-specific health trajectories, comparing clinical outcomes with and without anticoagulation, by representing individual sequences of health events and associated risks over time. I am a member of the MARKOV AF research group at the University of Turku, a multidisciplinary project focused on modeling clinical and subclinical atrial fibrillation related events to support improved understanding and decision-making in anticoagulation treatment.