Segmentation and binding in an oscillatory neural network

The authors present a model of coupled oscillating neural networks which can simultaneously perform segmentation and binding. The two networks have memory patterns which are independent of one another, yet the input contains them in pairs as, for instance, objects and attributes. When presented with a mixture of such pairs in a constant input, the activities of the corresponding patterns oscillate in a staggered fashion, exhibiting segmentation. Moreover, the phases of the pairs lock with each other, demonstrating binding. The underlying networks are feedback systems which are composed of excitatory neurons grouped into cell-assemblies representing the memories and inhibitory interneurons to which they are connected. The oscillatory nature comes about by dynamic thresholds which implement a fatigue effect for the excitatory neurons