Webinar (14)

Published at: 2017-04-25

Speaker: Dr. Ali Faqeeh

Postdoctoral Fellow
School of Informatics & Computing
Indiana University, Bloomington, USA

Title: Accuracy of percolation theories: Strong finite size effects in large modular networks

Coordinator: Dr. Fakhteh Ghanbarnejad

Time: Sunday, April 30, 2017, 16:30

If you are interested to attend, fill in the form here.


Many of the systems we observe in nature, in societies, or in infrastructures are in the form of a network of interacting units. This underlying network structure shapes the behavior of such systems and is an indispensable factor in maintaining their correct function. Likewise, the processes that operate on these systems are largely influenced by their network structure.


A class of widely studied processes in network science is referred to as percolation. Percolation processes investigate the alteration of network connectivity and also provide insights for a broad range of applications such as robustness of a network to failures or attacks, epidemics in contact processes, vaccination strategies, neuronal avalanches, and stability of gene regulatory networks.


In this talk, I introduce the message passing (MP) approach (AKA the belief propagation method) developed for percolation and several other processes. Then I present an approach to investigate and understand the sources of inaccuracy of such theories. I show that although the presence of short loops in networks is an important source of inaccuracy of tree-based theories such as MP approaches, this is not the only major source of error. I also introduce a phenomenon, not captured by previous studies, that in the bond percolation process causes inaccurate predictions for the MP approach and its degree-based reductions. I also describe a new method we developed which improves significantly over the predictions of the corresponding MP theory on real-world and synthetic networks.