DyANE: Dynamics-aware node embedding for temporal networks

摘要: Low-dimensional vector representations of network nodes have proven successful to feed graph data machine learning algorithms and improve performance across diverse tasks. Most the embedding techniques, however, been developed with goal achieving dense, low-dimensional encoding structure patterns. Here, we present a node technique aimed at providing feature vectors that are informative dynamical processes occurring over temporal networks-rather than itself-with enabling prediction tasks related evolution outcome these processes. We achieve this by using modified supra-adjacency representation networks building on standard techniques for static graphs based random-walks. show resulting useful paradigmatic processes, namely epidemic spreading empirical networks. In particular, illustrate our approach nodes' states in single instance process. how framing task as supervised multi-label classification allows us estimate entire system from partial sampling random times, potential impact nowcasting infectious disease dynamics.