Biological neural network models of associative memory (such as the seminal Hopfield Network, which earned
its creator the Physics Nobel prize last month) suffer from an artificial separation of storage and recall phases: the network's natural dynamics, which enable pattern completion at recall, are suspended during the storage phase.
I will present a cortical circuit mechanism which was previously implicated in a range of nonlinear computations in
sensory cortex, and show how it can solve the above problem. The mechanism involves a network of supralinear neurons, with strong recurrent excitatory connections,
which is stabilized against runaway excitation by recurrent inhibition. We show that such networks can naturally cross over from a regime of
memory storage to a regime of recall by simply scaling the external inputs. Finally, we propose that these regimes could correspond to
different phases of theta oscillations in the hippocampus, which have been previously linked to memory recall and storage.