The inference of the network of interactions between a set of variables
from their sampled activities frequencies and pairwise correlations.
Deducing functional interactions between stochastic variables from their observed
correlations is an ubiquitous problem in experimental science, e.g. in biology
(gene expressions, neural activity, genomics, proteomics), in finance, in sociology, ...
This important question is currently studied in various fields, such as statistical inference,
probability theory, machine learning, statistical physics.
Our work was motivated by the inference of functional couplings between retinal neurons from
multielectrodes recordings. We have developed an inference procedure, which looks, in a
controlled way, for the clusters of strongly interacting variables. The procedure was tested
also on synthetic data generated from Ising models.
We hope that our algorithm will be helpful to analyze the new generation of data
dealing with hundreds of interacting variables.
see the Publication list, Biophysics: Inverse Problems