Associative Memory with Biologically-Inspired Cell Assemblies

Associative memory is a central cognitive task. However, the actual biological architecture that supports this memory is not currently known, so simulating with biologically plausible neurons and topologies is an ideal mechanism to improve understanding of associative memory. Simulations of spiking networks that perform associative memory tasks lay the groundwork for utilizing biological neurons in cognitive tasks. Specifically, this paper explores simulations of spiking networks that perform associative memory tasks using Hebbian cell assemblies of neurons to represent nodes and synapses to represent associations. The first tasks use binary cell assemblies to perform two well-known cognitive tasks. Then the paper examines different topologies of excitatory neurons for basic assemblies and their performance as short-term memory. Lastly, larger assemblies are associated in 2/3 sets, where two active elements can retrieve the third. Future research is proposed to explore the potential use of these assemblies and associations in cognitive tasks. By investigating biologically and cognitively plausible topologies, learning, and neurons, simulations will lead to an improved understanding of neuro-cognition, and potentially to systems that surpass the brittleness and domain specificity of current AI systems.