Speaker
Yasser Roudi
(Nordita)
Description
Our observations from complex systems e.g. financial,
neuronal or gene regulatory networks are always done in the
presence of hidden nodes. This means that, we only see a
part of the network at any given time: we can record from
only a fraction of neurons in a cortical network, or have
access to data from only a part of the market. This raises
the question of what we can say about this hidden nodes, and
if their presence can be included on modeling the
high-throughput data collected from these complex systems.
Focusing on the kinetic Ising model as a prototypical
problem for learning and inference in kinetic models, in
this talk, I will describe how approximations based on
mean-field theory, belief propagation and replicas can be
used to recovering connections from partial observations in
this network.
This work is based on
Dunn and Roudi, PRE 2013
Battistin, Hertz, Tyrcha and Roudi 2014 arXiv 1412.1727
