IPN-batiment 100, Auditoriurm 15 Rue Georges Clemenceau - orsay Évènements

48.6981962.181773

Inverse inference in the asymmetric Ising model

In recent years new experimental methods have made possible the acquisition of an overwhelming amount of data for a number of biological systems such as assemblies of neurons, genes and proteins. Typically, these systems consist of a large number of interacting components and can be described by high dimensional models such as the well known Ising model from statistical physics. The nature of the data acquired from experiments makes necessary the development of methods that are able to infer the parameters of the model and thus predict the pattern of the interactions between the components of such systems. In this thesis I have studied the particular case of the Ising model with asymmetric interactions which is arguably the most relevant case when dealing with neural networks and could be generalized to fit to other biological systems as well. I will present a new mean-field inference method based on a simple application of the central limit theorem, able to infer exactly the parameters of the asymmetric Ising model from data in a computationally efficient way. I will also discuss some results of numerical simulations where the performance of our new method can be evaluated and compared with other existing methods. Finally, I will also show how the method can be better adapted to the case of sparse networks where, additionally, the amount of data used in the inference is low compared to the size of the system.