Seminar Details
2013-12-17 (14.00) : Low-rank Similarity based Extraction of Roles in Networks
At Euler building
Organized by Mathematical Engineering
Speaker :
Browet, Arnaud
Abstract :
In recent works, a lot of attention has been focused on the extraction of meaningful structures characterizing networks at aggregated levels.
Many researchers have proposed algorithms and methods to unfold community structures, i.e. groups of densely connected nodes, but little attention have been raised to extract more general structures known as roles. In this talk, we will present the problem of role modeling (a.k.a. Block modeling) and propose a pairwise similarity matrix that compares the flow pattern of each nodes in the network and allows the extraction of such role structures. Computing the exact similarity matrix being computationally difficult, we will present a low rank iterative scheme and demonstrate his convergence.