Abstract
Associating features of an environment allows an organism to learn from the past and predict the future. The binding of cortical cell assemblies, each representing different stimuli, by gamma-frequency synchronization is thought to be a method by which the brain associates units of information. It has been shown that individual cell assemblies, acting as gamma-frequency oscillators, represent associated stimuli by synchronizing their oscillations in the gamma band. Spike-time dependent plasticity has been shown to synchronize cells oscillating at these frequencies. We explore the possibility that such systems of oscillators can store associative memories by learning which groups are most often synchronized. We use a simple rate model to represent neural oscillator assemblies, and a mapping of the classical spike-time dependent plasticity (STDP) rule to rate models to represent plastic synapses. We find that cross-oscillator excitatory-to-excitatory connections between synchronized oscillators increase in weight, causing the oscillators to be more likely to be in-phase, and the same weights between non-synchronized oscillators decrease, causing the oscillators to be more likely to be out-of-phase. We also show that systems of multiple oscillators are able to perform pattern completion and separation.