A neuron based model for data storage and retrieval

Abstract

Memory is thought to be divided into two separate stores, one short term and one long term. The mechanism behind short term storage is significantly less well understood than that of long term storage and it is the storage of short term memories that is the focus of this work. A computational approach is taken to exploring a possible mechanism by which they are encoded and stored. This mechanism is comprised of a temporally patterned signal which exists persistently in a pre-existing network that has been stochastically constructed to support such behaviour. It is found that persistence of activity is achievable in a full neuronal model by introducing negative feedback in the form of a delay period in which a synapse may not fire. In addition, it is found that network cycles which could support storage of temporally patterned signals are achievable in simplified models constructed in analogy to the full model. Finally, some signs that storage may also be present in the full model are observed.