Federico Ricci Tersengh, Associate Professor, Dept of Physics, Universita Sapienza, Rome Chair Matthias Hennig |
Tue 26 May 2015, 11:00 - 12:00 |
Informatics Forum (IF-4.31/4.33) |
If you have a question about this talk, please contact: Mary-Clare Mackay (mmackay3)
TITLE
Improving mean-field approximations to better infer the interaction network
ABSTRACT
An interesting problem, common to several disciplines, is the one of inferring the interaction network among dynamical variables, given some configurations of these variables.
The most basic tool to solve this problem is the mean-field approximation.
I will discuss several attempts to improve over the naive mean-field approximation:
- its use in presence of a phase transition, that clusters the phase space;
- the use of high-order approximations, like TAP and Bethe;
- a modified approximation that is consistent with linear responses;
- a greedy algorithm (decimation) to impose sparseness in pseudo-likelihood methods.
http://chimera.roma1.infn.it/FEDERICO
There will be lunch provided after this seminar.