(V.B. Kazantsev, A.Yu. Simonov, A.S. Pimashkin) The project is to develop models of oscillatory associative memory in the networks of synaptically connected spiking neurons. Information (binary videoimages) is encoded in the form of definite distribution of phase clusters of spikes coming either in-phase or anti-phase with a base oscillatory rhythm. A set of information patterns is stored in the neuronal network due to appropriate choice of synaptic reversal potentials distributed according to a Hebbian learning rule. Responding to incoming stimulus the output layer of the network dynamically converges to one of the patterns stored in the memory. It corresponds to certain phase-shift relations of spiking relative to the base oscillation. The project includes the following tasks:
- Design of multi-layer network architecture to achieve maximal retrieval efficiency;
- Construction of optimal pattern "alphabet" to improve the storage capacity;
- The development of the associative memory networks with explicit axonal delays.
Fig. 1. Two-layer architecture of the associative memory network.
Fig. 2. Spiking phase cluster formation in the output layer to recognize the incoming information.