Publication
ICASSP 2023
Conference paper
MATRIX RESOLVENT EIGENEMBEDDINGS FOR DYNAMIC GRAPHS
Abstract
Eigenvector embeddings have been widely used to study graph properties in signal processing, mining, and learning tasks. However, if a graph is changing dynamically, these embeddings have to be recomputed. In this work we introduce a novel matrix resolvent expansion-based projection scheme to update eigenvector embeddings of dynamic graphs. The proposed method can tackle graph updates where both new vertices and edges are added, and its potential is illustrated via numerical tests on real data.