New publication in Nature Communications

Predicting the evolution of contagion dynamics is an important yet difficult problem, where simple mechanistic models, like mass-action models, can only take us so far. The paper entitled Deep learning of contagion dynamics on complex networks, newly published in Nature Communications and led by current Dynamica member Charles Murphy, alumni Edward Laurence and Dynamica leader Antoine Allard, introduces a complementary approach based on deep learning to learning effective dynamical models of epidemic spreading directly from observational data. The authors show that not only is their approach effective at learning from data and predicting dynamics on observed network structures, it is also possible to extrapolate it beyond the observed structure to extract emergent behaviors from the learned model.

Links to the full text and preprints are available on the publication page.