Local Detection of Communities by Attractor Neural-Network Dynamics

Community structure is a hallmark of a variety of real-world networks. Development of effective and efficient methods for detecting communities in networks has been a central issue of network science. Here we propose a method for detecting communities in networks. We have devised this method inspired by the cell assembly hypothesis, which has been one of the prevailing hypotheses in neuroscience. The cell assembly hypothesis states that neurons coding the same item tend to be mutually connected, thus forming a 'cell assembly'; memory recall of this item is associated with sustained activation of neurons belonging to the cell assembly. Here we compare communities to cell assemblies and examine community detection by use of the neural-network dynamics describing memory recall in the brain. To demonstrate the effectiveness of the proposed method, local detection of communities in synthetic benchmark networks and real social networks is examined. The community structure detected by our method is highly consistent with the correct community structure of these networks.