Associative Memory Models of Hippocampal Areas CA1 and CA3

Citation

Graham B, Cutsuridis V & Hunter R (2010) Associative Memory Models of Hippocampal Areas CA1 and CA3. In: Cutsuridis V, Graham B, Cobb S & Vida I (eds.) Hippocampal Microcircuits: A Computational Modeler's Resource Book. First ed. Springer Series in Computational Neuroscience, 5. New York, USA: Springer, pp. 459-494. http://www.springer.com/biomed/neuroscience/book/978-1-4419-0995-4; https://doi.org/10.1007/978-1-4419-0996-1_16

Abstract
The hippocampal regions CA3 and CA1 have long been proposed to be autoand heteroassociative memories, respectively (Marr, 1971; McNaughton and Morris, 1987; Treves and Rolls, 1994), for the storage of declarative information. An autoassociative memory is formed when a set of neurons are recurrently connected by modifiable synapses, whereas a heteroassociative memory is formed through modifiable connections from an input layer of neurons to an output layer. Associative memory storage simply requires a Hebbian strengthening of connections between neurons that are coactive (Amit, 1989; Hopfield, 1982; Willshaw et al., 1969). Recall proceeds from a cue activity pattern across neurons that is a partial or noisy version of a previously stored pattern. A suitable firing threshold on each neuron that receives input from already active neurons ensures that neural activity evolves towards the stored pattern. This may happen with only one or two updates of each neuron's activity. Accurate recall is obtainable provided not too many patterns have been stored, otherwise recall is poor, or even impossible. Network models of spiking neurons can be used to explore the dynamics of storage and recall in such memory networks. Here we introduce a recurrent network model based on hippocampal area CA3 and a feedforward network model for area CA1. Cells are simplified compartmental models with complex ion channel dynamics. In addition to pyramidal cells, one or more types of interneuron are present. We investigate, in particular, the roles of these interneurons in setting the appropriate threshold for memory recall.

People

Professor Bruce Graham

Emeritus Professor, Computing Science