IMM Publikationsdatabase

1
Learning latent structure in complex
networks Morten Mørup and Lars Kai
Hansen Cognitive Systems, DTU Informatics, Denmark
How does model flexibility affect identification
of latent structure?
1
Does latent structure (community detection)
modeling assist link prediction compared to
heuristic or non-parametric scoring methods?
2
To take degree distribution into account in the
latent modeling we propose the Link Density model
(LD) - an extension of the Mixed Membership
Stochastic Block Model (Airoldi et al, 2008).
2

• Most community detection approaches can be posed
  as a standard continuous optimization problem of
  what we define as the generalized Hamiltonian for
  graph clustering (GHGC)

We evaluated a variety of community detection
approaches and non-parametric methods in terms of
their ability to predict links (AUC score) on 3
synthetic and 11 benchmark complex networks
Community detection approach better than all
non-parametric methods
Non-parametric method better than all community
detection approaches
Proposed LD model best performing community
detection approach
We look very much forward to discuss these
results!