|Speaker :||Thomas Bonald|
|Time:||2:00 pm - 3:00 pm|
|Location:||LINCS / EIT Digital|
Learning from data strutured as a graph usually requires to embed this graph in some vector space of low dimension. The most popular technique relies on the spectral decomposition of the Laplacian. The focus of this talk will be on the “physics” of this spectral graph embedding, in various fields like thermodynamics, mechanics and electricity. These physical interpretations both suggest natural extensions to existing approaches and shed light on the crucial question of the choice of the Laplacian.
Part of this work was presented at the Allerton conference:
T. Bonald, A. Hollocou, M. Lelarge, Weighted Spectral Embedding of Graphs, Allerton conterence, 2018